CA2772082A1 - Generating a reference set for use during document review - Google Patents

Abstract

A system (10) and method (50) for providing generating reference sets (14b) for use during document review is provided. A collection of unclassified documents (14a) is obtained. Selection criteria (61) are applied (53) to the document collection and those unclassified documents (14a) that satisfy the selection criteria (61) are selected as reference set candidates. A classification code is assigned (55) to each reference set candidate. A reference set (14b) is formed (57) from the classified reference set candidates. The reference set (14b) is quality controlled and shared between one or more users.

Description

GENERATING A REFERENCE SET FOR USE DURING DOCUMENT REVIEW

TECHNICAL FIELD
The invention relates in general to information eta ic_~ 1 f ay , s e i t cf ll , cr a system :mad method for generating a a .ference sot for use during document review, BACKGROUND ART
Document review is an acti ity frequently undertaken in the legal field during the discover- phase of.litigaation. Typically, document classification requires reviewers to assess the rc.le ~anace of doca.aaa e its to a particulaar topic as aean init ial stela.
fit?cart ent e iews c:ta be conducted manually by human reviewers: automatically by a .machine, or by a combination of human :reviewers and a machine.
Generally, trained reviewers analyze documents and provide a recommendation for elassif vrng each document: in regards to the particular legal issue being litigated.. set of exemplar d.ocumeaats is provided. to the reviewer aas a guide ft-sr classifying, the documents. The exemplar documeaats are each previously classified with a particular code relevant to the legal issue, such as responsive:," "non-responsive," and "'privileged." Based on the exemplar doctuncrrts, the human revicwcrs or machine can idcrrtifv> documents that are similar to one or 24 more of the e ernplar documents and assign the code of the exemplar docuirteaat to the uncoded documents.
The set of exemplar documents selected for document review can dictate results of the review. A cohesive representative exemplar set can produce accurately coded documents, while effects of inaccurately coded documents can be detrirnentall to a legal proceeding, For example, a "privileged" document contains intortriation that is protected by a privilege, meanings that Elie document should not be disclosed to an opposing party. Disclosing a `,privileged'' document can result in tm unintentional waiver of privilege to the suihjec.t matter, The prior art focuses on document classification and general) assumes that exemplar documents are already defined and exist as a reference set for use in claassi#~-ing document. Such classification can benefit from having, better reference sets geiiea:ated to increase the accuracy of classified documents.
Tlaum there remains a aced. for a s ysteirt and method for generating a set oi exem afar documents that are cohesive: and W hich can serve as an accurate and efficient exaarupie for use in c:laassifvroar docutrierits.
-1 r DISC.LOSU=RE OF THE EN 4 f NTION
A system and nietlt.od for providing gene..rat.ing reference sets for use during document review is provided. A collection of unclassified documents is obtained.
Selection criteria are applied to the document cr_fillection and those unclassified docui eats that satisfy the selection criteria are selected. as reference set candidates. A classification code is assigned to each reference set ca.ndidite..A reference set is formed from the classified reference set candidates.
The reference set is quality controlled and shared between one or r tore.
users.
A further enrbofidument provides a method. for generating are rcnce set via clustering. A
collection of doctirnents is obtaineel. The documents are grouped into clusters of documents.
One or more of the documents are selected from at least one cluster as reference set candidates.
A classrf catioar code is asst ned to each of the reference set candidates.
The ckissffiod reference set candidates are grouped. as the. reference set.
A still further embodiment provides a r rethod for generating a reference set via seed docurrm tints. A collection of documents is obtained. One or z core s ed documents are i enntil ed.
The seed documents are compared to the document collection. Those documents that are similar to the seed documents are identified as reference set candidates. A size threshold is applied to the reference set candidates and the reference set candidates are grouped as the reference set when the size threshold is satisfied.
An even further embodiment provides a method for ~?eneratin a training set for Use 2Ã3 during document review. Classification codes are assigned to a set of documents. Further-classification codes assigned to the same set of documents are received. The classification code for at least one document is compared. with the further classification code for that doctunent. A
determination as to whether a disagreement exists between the assigned classification code and the further classification code is made for at least one document. Those documents with disagreeing classification codes are identified as training; set candidates. A
stop threshold is, applied tot the training set candidates and the training set candidates are grouped as a training set when the stop threshold Is satisfied, Still other embodiments of the present invention will become readily apparent to dose skilled in the art f:ron:t the fallowing detailed description, wherein are.
described e bodiments by 30 wav of illustrating the best erode contemplated for care ing out the inversion. As will be realized, the invention is capable of other and different embodiments and its several details are capable of modifications in various obvious respects, all without departing from the spirit and the scope of the present invention. Accordingly, the drawings and detailed description are to N, regarded as illustrative in nature and. not as restrictive.

DESCRIPTION O1 I'HE DRAWINGS
FIGURE 1 s a block diagram showing r system for generating r reference set for use during doeument review, in accordance with one embodiment.
FIGURE 2 is a flow diagram showing a method for generating a reference sot for use durimY document.re view, i:n accordance with one embodiment.
FIGURE 3 s a data flow diagram showing examples of the selection criteria of FIGURE 2, FIG ICI s a flow diagram showing, by way of example, a method for generating a reference set via.hier irchic it clustering.
I f} 1161. RE 5 is a flows diagram showing, b `ay of example, a iriethod for gcneratirig a reference sot via iterative clustering.
FIGURE 6 is a flow diagram showing, by way of example, a method f "rr generating a reference set via document. seeding.
FIGURE 7 is a flow diagram showia , by way of example, a r ethod for f er er itinig a refirencc set via random sampling.
FIGURE 8 s a flow diagram showing, by way, of example, a method for 4generating a reference set via user assisted niean&
FIGURE 9 is a flow diagram showing, by way of example, ,a method for generating r reference set Via active learn:im.
2Ã3 FIGURE I () is a flow diagram showing ~~ ati of e xarrrple ti method for gener<rting a traiirirr~: set.

l L`l .?~"T If 3`
BEST MODE FOR CARRYING OUT THE
Reference documents are each associated with a classification code and are selected as exemplar documents or a "reference set,` to assist human reviewers or a machine to identify and code unclassified clocurier ts. The quality of a reference sot can dictate the results of it document review project and an underlying legal proceeding or other activity. Use ofa noicohesive or "hid" reference set can provide inaccurately coded doeu:rrrerrts and could rfegativeb", affect a pead.im-, legal issue during, for Instance, litigation. Generally, reference sets should be cohesive or a particular issue or topic and provide accurate guidance to classifying dc)cunionts.
Cohesive r'efer'ence set. Ferieratioil requires a support environment to review, analyze., and select appropriate documents for inclusion in the reference set. FIGURE I is a block diagram showing a systerir for generating a referenc set for use in classift,ing docurrteuts, in accordance with one embodi:trietit. By way of illustration, the system 10 operates iii it distributed compoitii-i<
e.nvironme..nt, including "CIOUd environments," which include a plurality of systems and sources.

A backend server 1.1 is coupled. to a storage device 13, a d ttaLlase 3() for m-13 tataaia-aing? Inform 160 11 about the document a, and a lookup dartaabza ae 38 for storing many-to-many mappings 39 between documents and document features, such as concepts. The storage device 13 stores docurrients 14a and reference sets l4 b. The documents 14a cart include uncodod or "tanclassified"
documents and coded or "c.lassif e " documents, in the font of structured or unstructured data.
Hereinafter, the terms "classified" and "coded" are used interchangeably with the same intended moaning, unless otherwise indicated, The uarcodod and coded documents can he related to one or more. topics or legall issues.
Uncoded documents are analyzed and assigned a classification code, during a document review..
while coded documents that have been previously reviewed and associated with a classification code. The storage device. 13 also stores reference documents 14b, which together form a reference. sot of trusted and known results for use in guiding document classification. A set of reference documents can he hand-selocted or automatically selected, as discussed irn]`ra.
Reference sets can be generated for one or more topics or legal Issues, as well as for any other data to be organized and classified. For instance:., the topic can include data regarding a person, place, or object. In one embodiment, the reference set can be generated for a legal proceeding based on a filed complaint or other court or administrative filing or stibriussion.
Documents in the reference set 14h are each associated with an assi ned classification code and can highlight important information for the current topic, or legal issue. A
reference set can include reference documents with different classification codes, or the sane classification code, Core reference documents most clearly exhibit the particular topic or legal matter, whereas boundary condition reference documents include informaation similar to the core reference documents, but which are different enough to require ass ignment of a different classification code.
Once generated, the reference set can be used as a guide for classi irag encoded documents, such as described in commonly-assiggned U.S. Patent Application Serial No.
1 /833,81Ã1, entitled "System and Method for Displaying Relationships Between Electronically Stored information to Provide Classification Suggestions via. Ãncl-sion, "
filed July 9, 2010, pending. U.S. Patent Application Serial No. 12,83' ,872, entitled System and Method for Displaying Relationships Bets een Electronically Stored information to Provide Classification Suggestions via lajectio.n,," filed July 9, 2010, pending U.S. Patent .A.pplic<ation Serial No.
12"833,880, entitled "System and Method for Displaying Relationships Between Electronically Stored Information to Provide Classification Suggestions via Nearest Neighbor," filed July 9, 2010, pendin4g and U.S. Patent Application Serial No. 12'833,769, entitled "System and Method fbr- Proti idi:n a C.las'sificatiorr Srr ? estion for l lccironicaliy tor'ed .lrrfs rrnatis rl, filed on July 9, 201Ã1, pending, the disclosures of which are incorporated by reference.
In a further embodiment, a reference set can also be generated based on features associated with the docrtrrreart, The feature reference.. set can be, used to identify encoded.
documents associated with the reference set features and provide classification suggestions, such as described in commonly-assigned U.S. Patent Application Serial No.
121'844,810, entitled -System and Method for Displaying Relationships Between Concepts to Pro fide..
Classification Suggestions via Inclusion," filed July 2^17, 20111, pending; U.S. Patent Application Serial No.
12/844,792), entitled ""System and Method for Displaying Relationships Between Concepts to Provide Classification Suggestions via Injection filed July 27, 2010, pending;
U.S. Patent:
Application Serial No. 12 ,'844,813, entitled 'System and Method for Displaying Relationships Between Concepts to Provide. Classification Suggestions via Nearest Neighbor,"
led July 2'17 , 2010, pending; and U.S. Patent Application Serial No. 12,844,785, entitled "System and Method for Providing a { lassi icatiort Str =gs sÃion tor- Concepts," filed July 27, 2010, pendlug disclosures of which are incorporated by reference.
'1'he hackend server l 1 is also coupled to an intranetwork 221 and executes a workbench suite :31 for providing a user interfhce framework for autom ated document r , nagcral en t, processing, analysis, and etassific ation. In a further embodiment, the backend server l 1 can be accessed via an internetw rk 22. The workbench software suite 31 includes a document mapper 32 that includes a cirrstering engine 33, selector 34, classifier 35, and display- generator 36. Other workbench suite modules are possible. In a further embodiment. the clustering engine, selector, classifier, and display generator can he provided independently of the document mapper.
The clustering engine 33 perfbrmrts efficient document scoring and e usterinFg of encoded documents and reference docun-cents, such as described in commonly-assigned T
U.S. Patent No.
7,614:0,313, issued on October 27, 2009, the disclosure of which is incorporated by reference.
The encoded documents l.4a can be grouped into clusters and one or more documents can he selected from at least one cluster to fora reference set candidates, as f rr-ther discussed below in detail with r :.:ference to FIGURES 4 and 5. The clusters can be organized along vectors, known as spitzes, based on a sirnlarity of the clusters. The selector 34 applies predetermined criteria to a set of documents to identify candidates for inclusion in a reference set, as discussed iinfa, The classifier 35 provides a machine-generated classification code suggestion and confidence level for coding of selected uncoded documents.
The display generator 36 arranges the clusters and spines in thematic neighborhood relationships in a tcvo-di:n ensional visual display space. Once (generated, the visual display space is transr: utted to a work client 12 by the backend server I 1 vira the document im-1pper 32 for presenting to a human reviewer. The reviewer can include an iardividual person who is as-signed to review and classify one or more uncodcd documents by designating a code.
Other types of 11 .' rev=iewer$ are. Possible., including ~aaclairre iar:aplenreratecl reviewers.
The document .mapper 32 operates on encoded documents 141, which can be retrieved from the storage 13, as well as from a plurality of local and remote sources.
As well, the local -nd remote sources can also store the reference documents 14b. The local sources include documents 17 maintained in a storage device 16 coupled to a local server 15 and documents 2Ã3 maintained in a storage device 19 coupled to a local client IS. The local server 15 and local client 18 are interconnected to the bracketed server 11 and the work client 12 over an intranetwork 21. In addition, the document mapper 32 can identify- and retriev=e documents from remote sources-s over an Internees pork 22, including the. Internet, through a gatoway 23 interfaced to tile intranetwork 21. The remote sources include documents 26 maintained in a storage device 25 coupled to a remote server 24 and documents 29 mrtaintained in :t storage device 28 coupled to a remote client 27. Other document sources, either local or remote, are possible, The individual documents 1 Ala, 14b.17. 2i} 26, 23 include all forms and types of structured and unstructured data, i 3cludin4g electronic a ressaa e stores, word processing 11 documents, electronic mail (email) fbldcrs. Web pa-ges, and graphical or multimedia data.
Notwithstandi:ng), the documents could. be in the form of structurally organized data, such as stored in ,a spreadsheet or database.
In one crubodiment, the individual documents 14a, l 4b., 17, 20, 26, 29 include electronic message folders storing email and attachments, such as nraintamed. by the Outlook and Windows Live Mail prod ucts, licensed by Microsoft Corporation, Redmond, WA. The database can be. an SQL-based relational database, such as the Oracle database management system, Release 1.1, licensed by Oracle Corporation, Redwood Shores, Cry, Further, the individual documents 17, 20.26, 29 can b; stored in it "cloud,"' such as in Windows Live H.ot iail..licensed by Microsoft Corporation . Redmond, WA. Additionally, the individual documents 17, 20, 26, 29 include uncodcd documents and r :,ference documents.
The system 10 includes individual computer systerns, such as the backczrd server 1I, work serer 12, ser .vr15, client 1St rcatroÃe server ? and remote client'.".
"Tee indi\idual computer systems are general purpose, programmed digital computing devices that have a central processing unit (CPU), random access .memory (RA N,1), non-volatile secondary storagc, such as a hard drive or CD ROM drive, network interfaces, and peripheral devices, including user interfacing means, such as a keyboard and display. Program code, including software progrims, and data are loaded into the RANI. for execution and processing by the CTIU and resuits a.re generated for display, output, transmittal, or storage.
Reference set candidates selected for inclusion in a reference set are identified using selection criteria, Which cara reduce the number of documents for selection, Mir- RE 2 is a flow diagram showing a method for generating a reference set for use in document res,iew, in accordance with one embodiment. A collection of documents is obtained (block Si). The collection of documents can include encoded documents selected from a current topic or legal matter, previously coded documents selected from a related topic or legal matter. or pseudo documents. Pseudo documents are created rasing knowledge obtained by a person familiar with the issue or topic that is converted into a document. For example., a rev ieiNver who participated in a verbal conversation with al itsgaart or other party during which specifics of a lawsuit were..
discussed could create a pseudo document based on the verbal conversation, A
pseudo document. can exist electronically or in hardcopy form. In one embodiment, the pseudo docturm.it is created specifically for use dining the document review. Other types of document collections are possible.
Filter criteria are optionally applied to the document collection to identify a subset of documents (block 52) for generating the reference set. The filter criteria can he based on metadata associated with the documents, including date, file, folder, custodian, or content. Other filter criteria are possible:. In one example, a filter criteria could he defined as "all documents created after 1997:" and thus, all documents that satisfy; the filter criteria are selected as a subset of the. document collection.
The filter criteria can be used to reduce. the number of documents in the collection.
Subsequently, selection criteria are applied to the document subset (block.
53) to identify those documents that satisfy the selection criteria as candidates (block 54) for inclusion in the.
reference set. The selection criteria can include clustering, feature identification, assignments or random selection, and are discussed in detail below with reference to FIGURE
3. A candidate decision is applied (block 55} to the reference set candidates to identif.' the reference candidates for potential inclusion in the reference set (block 57). During, the candidate decision, the reference set candidates are analyzed and a classification code is assigned to each reference set candidate. A human reviewer or machine can assign the classification codes to the reference set candidates based on features of each candidate. The features include pieces of information that described the document candidate, such as entities, metadata, and summaries, as well as other information. Coding instructions guide the review er or machine to assign the correct classification code using; the features of the reference set candidates. The coding instructions can be provided by a r-etiie per, a supervisor, a law- firm, a party to a legal proceeding, or a machine.
Other sources of the coding instructions are possible:.
Also, a determination as to whether that reference set candidate is a writable candidate for including in. the reference set is Wade. Once the reference set candidates are coded, each candidate is analy zed to ensure that candidates selected for the reference set cover- or "span" the largest area of feature space provided by the document collection- inr one embodiment, the candidates that are most dissin-tilar from all the other candidates are selected as the reference set.
A first reference set candidate is selected and placed in a list. The ren-taining reference sot candidates are compared to the first reference set candidate in the list and the candidate most dissimilar- to all the listed candidates is also added to the list. The process continues until all the dissimilar candrdates have been. identified or other stop criteria have been satisfied. The stop criteria can. include a predetermined number of dissimilar reference sot criteria, -,ill the candidates have been reviewed, or a measure of the most dissimilar document fails to satisfy a dissimilarity threshold. Identifying dissimilar documents is discussed in the paper, Sean M.
Mc,Nec. "Meeting User Information Needs in Recommender Systems". Ph.D. Dissertation, University of Minnesota-Twin. Cities. June 2006, which is hereby incorporated by reference.
Other stop criteria are; possible-1-l.owcver, refinement (block 56) of the reference set candidates can optionally occur prior to selection of the reference set The refinement assists in narrowing the :number of reference set candidates used to generate a reference. set of a particular size or Other criteria. If refinement rs to occur, further selection criteria are applied (block 53) to the reference set candidates and a further iteration of the process steps coccurs. Each iteration can involve different selection criteria. For example, clustering criteria can be applied during a first pass and random sampling can be applied during a second pass to identify reference set candidates for inclusion in the reference set.
In a f rrfher embodiment, features can be used to identify documents for inclusion in a reference set. A collection of documents is obtained and features are identified from the document collection, The features can be optionally filtered to reduce the feature set and subsequently, selection criteria can be applied to the features, The features that satisfy the selection criteria are selected as reference set candidate features. A
candidate decision, including assigning classification codes to each of the reference set candidate features, is applied, Refinement of the classified reference set candidate features is optionally applied to broaden or narrow the reference set candidate features for inclusion in the ref rence set, The refinement can include applying farther selection criteria to reference set documents during a second iteration.

.Alternatively, the selection crirerm can first be applied to docutricnts and in a further iterations the selection criteria are. ap iied to features from the docun: eats.
Subsequently, documen s associated with the .referrenc.e set candidate features are grouped as the reference set, The c rndidate criteria can be appiied. to a document set to identity reference set candidates for potential inclusion in the reference set. FIGURE 3 is a data flow diagram 60 shca~i ins examples of the selection criteria of FIGURE 2. The selection criteria 61 include duster n 62, features 63, assignments 64, document seeding t??, and random sarripling fi(i.
Other selection criteria are possible. ClusÃe_ting, 62 includes grouping documents by similarity and subsequently selecting documents from one or more of the clusters .A
number of documents to be selected can be predetermined by a reviewer or machines; as further described below with reference. to FIGURES 4 and 5. Features 63 include rraeÃadata about the dr cements, including nouns, noun Phrases, length of document, "'i`re" and "From" fields, date, complexity of sentence structure, and concepts. Assignments 64 include a subset of documents selected from a, larger collection of n rcoded document to be reviewed. The assignments can be generated based on assignment criteria, such as contertà size, or number- of reviewers. Other (atures, assignments, and assignment criteria are possible.
Document seeding 65 includes selecting one or more seed documents and idei-ififvInQ
documents similar to the seed documents from a larger collection of documents as reference set candidates. I ocument seedi:n is further discussed below in detail with reference to FIGURE 6.
Random samplinrn 66 includes randomly selecting documents from a l rrger Collection of documents as .reference set candidates. Random sampling is further discussed below in detail with reference to MiU RE 7.
The process for generating a reference set can be iterative and each pass through the process can use different selection criteria, as described above with reference to FI URE 2.
Alternatively, a single pass through the process using only one selection criteria to generate a cohesive reference set is also possible. Use of the clustering selection criteria can idei-itifyand group documents by similarity. .FIG RE 4 is a flow diagram showing, by way of example, a method forgenezating a reference set via hierarchical clustering. A collection of documents is obtained (block 7 l) arnd filter criteria can option ally be applied to reduce a number of the documents (block 72), The documents are. then. clustered (block 73) to generate a hierarchical tree via hierarchical clustering. Hierarchical clustering, including g flc?rr erati e: or derisive clustering, can be used to generate the clusters of documens, which can be used to .identify a set of reference documents having r. particular predetermined size. During agglun erative clustering, each document ns assigned t0 a cluster and similar clusters are combined to generate the E

hierarchical tree. : icanwh te, during, 4 des si e clustering, all the documents are grouped into a single cluster and subsequently divided to generate the hierarchical tree.
The clusters of the hierarchical tree can he traversed (block 74) to identify, n docnrtxnents as reference set candidates (block ^175). The n-docr. ments can be predetermined by a user or a machine.. In one embodiment, the n--documents are influential documents, meaning that a decision :made for the n-document, such as the assignment of a classification code, can be Propagated to other sirnnilar documents. -using irnfluent f l documents can improve the speed and classification consistency of a document review. To obtain the r=-docnr_ments, n-clusters can he identified during the traversal of the hierarchical tree and one document from each of the identified clusters can be selected, The single document selected from each duster can be the document closest to the cluster center or a notlner doctrrnnctnfs. Other values of n are, possible., such as rv 2.. For exatnnt le., n/2. clusters are identified during the traversal and two documents are selected from each identified cluster. In one etribodir'=nnent, the selected documents are the document closest to the cluster center and the document furthest from the cluster center-. However, other documents can be selected, such as randomly picked documents, Once identified, the :reference set candidates are analyzed and a candidate decision is made (block 76). During the analysis, a classification code is assigned to each reference set candidate and a determination of whether that reference set candidate is appropriate for the reference set is made. If one or:more of the reference set candidates are not sufficient for the reference set:. ref inetnrent of the reference set candidates may optionally occur (block 77) by rec.lr:tsterinnn the reference set candidates (block 73). Refinement can.
include changing input parameters of the. clustering process and. then reclustering the documents, changing the document collection by filtering different documents, or selecting a different subset of n-documents from the clusters. Other types of and. processes for refinement are possible- The refinement assists in narrowing the number of .reference set candidates to generate a reference set of a particular size during Which reference set candidates can he added or removed. One or more of the reference set candidates are grouped to form the reference set (block 78). The size of the reference set can be predetermined by a human reviewer or a machine.
0 In a further embodinnent, features can he used to identify documents for inclusion in a reference set. A collection of documents is obtained and fertures from the documents are identified. Filter criteria can optionally he applied to the features to reduce the number of potential documents for inclusion in the reference set. The features are then grouped into clusters. Which are traversed to identif n--features as reference set candidate features..A

candidate decision, including the assi<grrrraeut of classification codes, is applied to each of the reference set candidate features and refinement of the features is optional.
Documents associated with the classified reference set candidate features are then grouped as the reference set.
Iterative clustering is a specific type of hierarchical clustering that provides a reference set of documents havin4k an approxinnate size. FIGURE 5 is a flow diagram showing, by ways of example, a method for generating a reference set via iterative clustering. A
collection of documents is obtained. (block $1). The documents can he optionally divided into assignments (block 8'), or groups ofdocume.nts, based on document chara,cte.rrsties, including mctadata about the document. In cneral, exist.in knowle&e about the document is used to generate the 1:0 assi rarrrerrts Other processes #trr g rrer,rtirrt tlre . assiy~rr ~i rrts are à ossible, In on. r~ bodiment:;
<rttaclrtractrts to the doc113n1ent can be included in the. s rnme assignment as the. docurnrent, and in an alternative embodiment, the attachments are identified. and sot aside for review or assigned to a separate assignment. The documents are then grouped into clusters (block 83).
One or more documents can be selected from the clusters as reference set candidates (block 84). in one embodiment, two documents are selected, including the document closest to the cluster center and the document closest to the edge of the cluster. The document closest to the center provides information regarding the center of the cluster, While the outer document provides information regarding the edge of the cl ustcr. Other numbers and types of documents can be selected.
The selected documents are them anal zed to determine whether a sufficient number- of documents have been identified as reference set mmdidates (block 85). The number of documents can be based on a predefined value, threshold, or bounded ran :e selected by a reviewer or a machine, If a sufficient number of reference set candidates are not identified, further cluster ing (block S3) is peri%rrmed on the reference set candidates until a sufficient number of reference set candidates exists. However, if a sufficient number of reference set candidates are identifed, the candidates are analyzed and a candidate decision is made (block 86). For "'11111)'C7 a threshold can define a desired number- of documents for inclusion in the reference set.lf the number of reference set candidates is equal to or below the threshold, those candidates are further analyzed, whereas if the r rumber of reference set candidates is above the threshold, further clustering is performed until the number of candidates is suihcient. In a further c <rsrnlnlt_, a bounded range, having an, nipper limitation and a lower lirrnitation, is determined and if the number of reference set candidates falls within the bounded range, those reference set candidates are further analyzed.
The candidate decision includes coding of the documents and a determination as to whether each reference set candidate is a good candidate: for inclusion in the reference set. The rcil-coded reference set candidates ffrrri the reference set (block 87}. Once fbrn e.d, the re"fere"nce set can be used as a group of exemplar documents to classify uncoded documents.
In a further embodiment, features can be used to identify docti-me Its for ncinsinn in the reference sot.. A collection of doctrrnents is obtained and features are identified within the documents. The features can optionafl y be divided into one or more assignments. The features are then grouped into Clusters- and at least one ft atu:re is selected from one or more of the clusters. The selected features are compared with a predetermined number of documents for inclusion in the reference set_ If the predetermined number is not satisfied..
further clustering, is performed on the features to increase or reduce the number of features.
However, if satisfied, the selected features are assigned classification codes. Refinement of the classified features is optional. Subsequently, documents associated with the classified features are identified and.
grouped as the reference set.
The selection criteria used to identify reference set mndidates can include document seeding, which also groups similar documents, FIGURE 6 is a flog w dial ram shhtowirr , by way of example, a rr::tetlhtod for generating a reference set via document seeding. A
collection of documents is obtained (block 91}. The collection of documents includes unmarked documents .related to a topic, legal matter, or other theme or purpose. The documents can be optionally grouped into individual asst ;mrnents (block 92). One or more seed docun eats are identified (block 93). The seed documents are considered to be important to the topic or legal matter and can include documents identified from the current matter, documents identified from a previous m>rtter, or pseudo documents.
The seed documents from the current arse can include the complaint filed in a legal proceeding for which documents are to be classified or oÃher documents, as explained suu p raa.
A.lÃernatively, the seed documents can he quickly identified using a keyword search or 1 rro lc.d4yc obtained f'gar:n a reviewer. In a further embodiment, the seed documents can be identified as reference set candidates identified in a first pass through the process described above with reference to FIGURE 1 The seed documents frorrr a previous related matter can include one or rrrore of the reference documents from the reference set generated for the previous matter:. The pseudo documents use knowledge from a reviewer or other user, such as a part),, to a lawsuit, as described above with reference to MGURE 12.
The seed documents are then applied to the document collection or at least one of the assignments and documents similar to the seed documents are identified as reference set candidates (block 94). In a farther embodiment, dissimilar documents can be identified. as reference set candidates. In yet a further embodiment, the similar and dissimilar documents can be combined to forno the seed documents. The similar and dissimilar documents can be identified. using criteria, including document injection, linear search, and index took tip.
floe=ever, other reference set selection criteria are possible.
The number of ref re_nce_ set candidates are analyzed to d.eterrnine whether there are a sufficient number of candidates (block 95 The number of candidates can be predetermined and selected by a reviewer or machine. If a sufficient number of reference set candidates exist, the reference set candidates form the ret:crence set (block 97). However, if the number of reference set candidates is not sufficient, such as too large.. refinement of the candidates is performed to remove one or More reference candidates from he, set (block 96.). Large reference sets can affect the perfor -nance and outcome of document classification, The refinement assists in narrowing the number ofrethrence. set candidates to generate a reference set of a particular size, If refinement is to occur, further selection criteria are. applied. to the reference set candidates. For example, if too many reference set candidates are Identified. the candidate set can he narrowed to re mro e COT 11111011 or closely related doctaments, while leaving the most important or representative document in the candidate set. The common or closely related documents can be identified as described in coc3ratro:nly-assigned US- Patent No. 6x~ 745,197, entitled "System and Method for Efficiently Processing Messages Stored in Multiple Messa4ge Store:s," issued on. June 1, 2004, and U.S. Patent No. 682081- entitled "System and Method for Evaluating a Structured Message Store for Message Redundancy," issued oar :November 16, 2004, the disclosures of which are incorporated by reference. Additionally, the common or closely related documents can to identified based on influential documents, which are, described a bovc with reference to FIGURE. 4, or outer measures of document similarity, After the candidate set has been refined, the remaining reference set candidates t`orrrr the rcfcrerrce set (block 97).
In a further embod.inrenÃ, features can be used to identify documents for incursion in the reference set. A collection ofdoccnnents is obtained and features from the documents are identified. The features arc optionally divided into assignments. Seed features are identified and.
applied to the. identified features. The features similar to the. seed features are identified as reference set candidate features and the similar features are analvred to determine whether a sufficient number of reference set candidate features are. identified. If not, refinement can occur to increase or decrease the number of reference set candidate features until a sufficient number exists. if sc?. document associated s .itlr thi reference set candidate #e turi s <r.re icl ratified rntl grouped as the reference set.
Random sampling can also be used as select on criteria to identify reference set candidates. FIGURE 7 is a flow diagram showing. by way of example, a me hocl for generating as :reference set Via random sampling. A collection of docranaents is obtained (block 101), as described above with .refe..rence to FI LIRI 2. The documents are then grouped into categories (block 102) based on nictadata about the doccuments. The n ctadata can ind ude date.: ile, folder, fields, and structure. Other anetadata types and groupings are possible.
Docuirtent identihcaataon values are assig7ned (block 103) to each of the documents in the collection.
The identification values can include letters, :numbbers, symbols o:r Color coding, as well as other Values, and can be haaraaan readfable or taaaachine readable.. t andotaa taaaach ne generaata_ar or a hraaaaan re e er cara assign the identification values to the documents. Subsequently, the documents are randomly ordered into a list (block 104) and the first n-documents are selected from the list as reference candidates (block 105). In a further embodiment, the document identification values are Provided to a random number generator, which randomly selects n document identification values. The documents -ssociaated with the selected identification values are then selected. as the reference set candidates. The n amber of n-documents can be determined by a hta.nan reviewer, user-, or machine. The value of n dictates the size of the reference set. The reference candidates are then coded (block 106) and grouped as the reference set (block I07).
In a further s:mbodiaaaent. features or terms selected from the documents in the collection can be sampled. Features can include me:tadhata about the documents, including anotaras noun phrases, length of document, "To" and "From" fields, date, complexity of sentence structure, and concepts. Other features are possible.. Identification Values are, assigned to the features and a subset of the features or terms are. selected, as described supra.
SaubseiTuently, the subset of features is randomly ordered into a list and the first: n-features are selected as reference candidate features. The documents -ssociated with the selected reference candidate features are then grouped as the reference set. Alteratively, the number of n features can be randomly selected by a random number generator, which provides n-feature identification values.
The features associated with the selected n-feature identification values -are selected as reference candidate feaatu:res.
Reference sets for coding documents by a human reviewer or a machine can be the same set or a different set, Reference sets for human reviewers should be cohesive;
but need not be representative of a collection ofdocuta}ents since the reviewer is comparing uncoded documents to the reference documents and identifying the sinillar auncoded documents to assign a classification code. Zieanwhile, a reference or "training" set for classifiers should be representative of the collection of documents, so that the classifier can distinguish. between documents having different classification codes. FIGURE 8 is a flow diagram showing, by way of e .aammaple. a method 110 for generating a reference set with user assistance. A collection of documents associated with a topic or .lei zl issue is obtained (block I 11). A
:reviewer marks one or more of the documents in the collection by assigning a classification code (block 112).
Together; the classified documents can form an initial or candidate reference set, which can be subsequently tasted and refined. The reviewer can raridomly select the documents, receive review requests for particular documents by a classifier, or receive a p:rccletc:rnii.ned list of docrrnrcrrt.s f rr marl inl;. In one embodiment, the documents marke :l by the reviewer can be considered reference documents, which can be used to train a classifier..
While the reviewer is marking the docurnents, a machine classifier analyzes the coding decisions provided by the reviewer (block 113). The analysis of the coding decisions by the classifier can include one or more steps, which can occur sinru1'taneor#sly or sequentially, In one embod.inrent, the analysis process is a training or retraining of the classifier. Retraining of the classifier can occur when new information, such as documents or coding decisions are identified.
In a further embodiment, multiple classifiers are utilized. Thereafter, the classifier begins classifying documents (block 114) by automatically tssiignin ; classification codes to the documents. The classifier can be in classification based on factors, such as a predcterm "tiled number- of documents for review by the classifier, after a predetermined time period has passed., or after a predetermined number of documents in each classification category is reviewed- For instance, in one embodiment, the classifier can begin classifying documents after analyzing at least two documents coded by the reviewer. As the number of documents atralyzed by the classifier prior to classification increases, a confidence level associated with assigned classification codes by the classifier can increase.. The classification codes provided by the classifier are compared. (block 115) with the classification codes for the.
same documents provided by the reviewer to determine whether there is a disagreement bets.
een the assigned codes (block 116). For example, a disag 'cement exists when the reviewer assigns r classification code of privileged" to a document and the. classifier assigns the same document a classification code of "responsive ' If a disagreement does not exist (block 1.16), the classifier begins to automatically classify documents (block 118), l=lcrwever, if a disagreement exists (block 116), a degree of the disa.g-reerr}ezrt is analyzed to deterrrmrine whether the disagreement Rills below a predetermined threshold (block 117). The predetermined. threshold can. be measured using a percentage., bounded range, or value, as well as other measurements. l:n one embodime rt, the disagreement threshold is set as 9# ,''.;~ agreement, or alternativel'y as P/,'% disargree ment. In a further enrbodiment, the predetermined threshold is based on a number of agreed upon documents. For example, the threshold can require that the last 100 documents coded by the reviewer and the classifier be in agreement. In yet a further embodinmcnt, zero-defect testing can be used to determine. the threshold. defe:et, r,tat oe clis<art ecnaent iaa a reveling de ision, satrla as ata ncorasisterac iaa th e l<assifrc.rtt.icyar coefr assigned. An error rate for classification is determined based on the expected percentages that a 1 articular classification code will be assigned, as well as a co:nfdence level. The error rate can include a percenta(ge, number, or other value, A
collection of documents.israndomly sampled and marked by the reviewer and classifier. If a valrie_ of docuirtents with disagreed uport classif cation. codes exceeds the error rate, further training of the classifier is necessary. However, if the value ofdocLartents having a disagreement falls below he error rate, automated classification can begin.
If the disagreement value is below the threshold, the classifier begins to automatically classify documents {block 11 S). if no tlae res e.~~er continues to rat<ark docctaracarts fiDrat the collection set (block 112), the classifier analyzes the coding decisions (block 113), he. classifier marks documents Mock 114), and the classification codes are, compared (block 115) until the disagreement of the classification codes assigned by the classifier and the reviewer falls below the predetcnrtined threshold.
In one embodiment, the disagreed. upon documents can he selected and 4arouped.
as thr.e reference set. Alternatively, all documents marked. by the classifier can be included in the reference seat, such as the agreed and disagreed upon documents.
In a further embodiment, features can be used to identify documents for inclusion in the reference set. A collection of documents is obtained and features are identified from the collection. A reviewer marks one or more f eatures by assigning classification codes and provides the marked features to a classifier for analysis. After the analy-sis, the classifier also begins to assign classification codes to the features. The classification codes assigned by the reviewer and the classifier for a common feature are compared to determine whethera disagreement exists. If there is no disagreement, classification of the features becomes automated. However, if there is disagreement. a threshold is applied to determine whether the disagreement falls below threshold. If so, classification of the features becomes automated.
However, if not, further marking, of the e atur~ s, and analysis occtars.
Reference sets generated using hierarchical clustering, iterative clustering, random sampling, and. document seeding rely on the human reviewer for coding, of the reference documents. Flowevm a machine, such as a classifier, can also be trained to identify reference sets for use in. classifying documents. FIGURE 9 is a flow diagram showing, by way of exam l'.e, 4a metlitrd for gea crating a refire aace set: is ~actise le.rtaaririg, t set of coded docurtrea is is obtained (block 121). The set of documents can include a document seed set or a reference set, as Weil as other t =pes of docutraent sets. `f'he derctlrrler3t set can Ire obtained f rot1 pre~-iotls related topic or legal matter, as Well as from documents in the current matter. The coding of the document set can be performed by a human reviewer or a.machine. The document set can be used to train one or more class] fiers (block 122) to identify documents for inclusion in a reference set. The classifiers can he the same or different, inluding:nearest neighbor or Support Vector Machine Classifiers, as well as other types of classifiers. The classifiers review and mark a set of uncoded documents fort particular topic, legal matter, theme, or purpose by assigning a classification code (block 123) to each of the uncod.ed documents, The classification codes assigned by each classifier for the same document are compared (block 124) to deteniiine whether there is a disagreement in classification codes provided by the classifiers (block 1225), A
disa,a,rcement exists when one document is assigned diff=erent classification codes, if there is no disagreement, the ifiers continue to reiew and classify the rcracoclcd.
documents (block 123) until there are no encoded documents remaining. Otherwise, if there is a disagreement, the document is provided to a human rev ieWer for review rnd mruarkinl . The human reviewer provides a new classification code or confirms a classification code assigned by one of the classifiers (block 1?6). The classifiers that incorrectly :marked the document and reviewer trssin ned classification code (block 127) can he analyzed fir further training. For the classifiers that correctly marked the document (block 127), no additional training need occur. The documents receiving inconsistent classification codes by the classifiers form the reference set

2(1 (block 128). The reference set can then he used to train further classifiers for classifyen documents.
In a further embodiment, features can he analyzed to identify reference documents fair inclusion in a reference set. A collection of coded documents., such as a seed set or reference set, is obtained. The document set can he obtained from a previous related topic, legal matter, theme or purpose, as well as from documents in the current tt titter. Features within the document set are identified. The features can include lmetadata about the. documents, includin :nouns, noun phrases, length of document. to and from fields, date, complexity of sentence structure, and concepts. Other features are possible. The identified features are then classified by a human rev tower and used to train one or more classifiers. Once trained, the classifiers review :u further set of uncoded docunments, identify features within the further set of uncodcd. documents, and assign classification codes to the features. The classification codes assigned to a common feature by each classifier are compared to determine whether r discrepancy in the assigned classification code exists. If not, the classifiers continue to review and classify the features of the uncoiled. documents until no uneoded documents rermlain. If there rs a classification 1~

3 PCT/US2010/046557 disa44ree:n-ae:trt, the feature is provided to as .humin :reviewer for anal ysis and coding. The classification code is received from the user and used. to retrain the classifiers, which incorrectly coded the feature. l c?catarae.nts aasse~eiaa:tcd with th disats*aeecl aapcan fea turc s titre idea t.ific d aand.
grouped to firm the reference set.
Feature selection can be used to identify specific areas of two or more documents that are interesting based on the classic c ation. disagreement by highlighting or marking the areas of the docurr-tents containing the particular disagreed upon features, Documents or sections of docurr-tents can be considered interesting based on the classification disagrecu-tent because die document data is prorripting multiple classifications and should be further reviewed by a human reviewer, In yet a further embodiment, a combination of the reference documents identified by docurnent and the reference documents identified by features can be combined to create a sinule reference set of documents.
The reference set can be provided to a reviewer for use in irianually coding docuirieuts or can be provided to a classifier for automatically coding the docuirtents. 111 a further embodiment.
dif :brent reference sets can be used for providing to a reviewer and a classifier. FIGURE 1.0 is a flow diagram 130 shosw ing, by way of example. a method for generating as trairilrizg set fora classifier. A set of coded document, such as a reference set, is obtained (block 13.1). One or more classifiers can be trained (block 1:3 ) using the referearce set. The classifiers can be the saute or different, such as a nearest neighbor classifier or a Support Vector Machine classifier.
Other types of classifi,ers are possible, Oince trained, the classif-ters are each man over a cotmrmmmorn sample of f ssigntaac.trts to classify documents in that assignment (block 133). The classification codes assigned by each classifier are analyzed for the documents and a deten-nrnation of whether the classifiers disagree on a particular classification code is rriade (block 134). If there is no disagreement (block 134), the classifiers are run over further coca mon samples (block 13,3) of assignments until disagreed upon documents are identified..Howev-er, if there is disagreement between the classifiers on a document marking, the classified document in disagreement must then be reviewed (block 135) and identified as .s training set candidates..
further classification code. is assigned to the classified document in disagreement (block 137). The further classification code can be :assigned by a human reviewer or a machine, such as one of the classifiers or a different classifier. The classifiers can each be option tllyr updated (block 132) with the newly assi~Yned coode. The review and document coding can occur manually by a reviewer or automatically. The training set candidates arc then combined wvith the reference set (block 1311). A stop threshold is applied (block 138) to the combined ta;ainin4g set candidates and reference set to deten.rmrne whether each of the documents is appropriate for inclusion in the training set. The stop threshold can include a predeteri-nined training set size.., a breadth of the training, set candidates with respect to the feature space of the reference set, or the zero defect test. Other types of tests and processes fir determining the stopping threshold are possible. If the threshold is not satisfied, the classifiers are run over further assignments (block 133} for classifying and comparing. Otherwise, if satisfied, the combined training set candidates and reference set fbrlrr the training set (block 139), Once generated. the training set can he used. fi r automatic classification of dc)cuments, such as described above with reference to FIGURE S.
In a further enrhod.iment, features can he used to identify documents for inclusion in the reference set. A set of coded documents is obtained and features are identified from the coded documents. Classifiers are trained using the features and then run over a random sample- of features to assign classification codes to the features. The classification codes for a common feature are. compared to determine whether a disagreement exists. If not, further features can be classified. However, if so, the disagreed upon features are provided to a reviewer for firrther analysis. The reviewer can assign further classification codes to the features. which are grouped as training set candidate features. The: documents associated with the training set candidate features can he identified as training set candidates and combined with the coded documents. A
stop threshold is applied to deterimne whether cac:h of the documents is appropriate for inclusion in the reference set. If so, the training set candidates and coded documents are identified as the training set. However, if not, further coding of features is performed to identif training set candidates appropriate for inclusion in the reference set.
While the invention has been particularly shown. and deserrkred. as referenced to the embodiments thereof, those skilled. in the art will understand that the foregr_rirrg and other change c in form and detail rr a y be made therein without de. rung from the spirit and scope of the invention.

Claims (20)

CLAIMS:

1. A method (50) for generating a reference set (14b) for use during document review, comprising:
obtaining a collection of unclassified documents (14a);
applying selection criteria (61) to the collection and selecting those unclassified documents (14a) that satisfy the selection criteria (61) as reference set candidates;
assigning a classification code (55) to each reference set candidate; and forming a reference set (14b) from the classified reference set candidates, wherein the reference set (14b) comprises coded documents (14a) that are quality controlled and shared between one or more reviewers.

2. A method (50) according to Claim 1, further comprising:
refining the reference set (14b) by reducing a number of reference set candidates included in the reference set (14b).

3. A method (50) according to Claim 1, wherein the selection criteria (61) comprises random sampling (100), further comprising:
randomly ordering the unclassified documents (14a) into a list; and selecting a predetermined number (105) of the unclassified documents (14a) from the list as the reference set candidates.

4. A method (50) according to Claim 1, wherein the selection criteria (61) comprises random sampling (100), further comprising:
assigning document identification values to each of the unclassified documents (14a);
selecting one or more of the document identification values (105); and identifying the unclassified documents (14a) associated with the selected document values as the reference set candidates.

5. A system (10) for generating a reference set (14b) for use during document review, comprising:
a collection of unclassified documents (14a);

a selection module to apply selection criteria (61) to the collection and to select those unclassified documents (14a) that satisfy the selection criteria (61) as reference set candidates;
a classification module to assign a classification code (55) to each reference set candidate; and a data module to form a reference set (14b) from the classified reference set candidates, wherein the reference set (14b) comprises coded documents (14) that are quality controlled and shared between one or more reviewers.

6. A system (10) according to Claim 5, further comprising:
refining the reference set (14b) by reducing a number of reference set candidates included in the reference set (14b).

7. A system (10) according to Claim 5, wherein the selection criteria (61) comprises random sampling (100), further comprising:
randomly ordering the unclassified documents (14a) into a list; and selecting a predetermined number (105) of the unclassified documents (14a) from the fist as the reference set candidates.

8. A system (10) according to Claim 5, wherein the selection criteria (61) comprises random sampling (100), further comprising:
assigning document identification values to each of the unclassified documents (14a);
selecting one or more of the document identification values (105); and identifying the unclassified documents (14a) associated with the selected document identification values as the reference set candidates.

9. A method for generating a reference set (14b) via clustering, comprising:
obtaining (71, 81) a collection of documents (14a);
grouping the documents, (14a) into clusters of documents (14a), selecting (75, 84) one or more documents (14a) from at least one cluster as reference set candidates;

assigning (76, 86) a classification code to each of the reference set candidate and grouping the classified reference set candidates as the reference set (14b).

10. A method according to Claim 9, further comprising:
building a hierarchical tree of the clusters; and traversing (74) the hierarchical tree to identify the reference set candidates.

11. A method according to Claim 9, further comprising:
applying (85) a size threshold to the reference set candidates; and selecting the reference set candidates for inclusion in the reference set (14b) when the size threshold is satisfied.

12. A method according, to Claim 9, further comprising:
applying (85) a size threshold to the reference set candidates; and reclustering the reference set candidates when the size threshold is not satisfied until the size threshold is satisfied.

13. A method for a reference set (14b) via seed.
documents, comprising;
obtaining a collection of documents (14a);
identifying (93) one or more seed documents (14a);
comparing (94) the seed documents (14a) to the document collection and identifying those documents (14a) similar to the seed documents (14a) as reference set candidates;
applying (95) a size threshold to the reference set candidates, and grouping the reference set candidates as the reference set (14b) when the size threshold is satisfied.

14. A method according to Claim 13, further comprising:
refining (96) the reference set (14b) by reducing a number of reference set candidates included in the reference set (14b).

15. A method according to Claim 13, further comprising:

selecting the seed documents (14a) from at least one of a current document set and a previously defined document set.

16. A method according to Claim 13, further comprising:
determining the seed documents (14a) using a keyword search.

17. A method for generating a training set for use during document review, comprising:
assigning classification codes to a set of documents (14a);
receiving further classification codes assigned to the same set of documents (14a);
comparing (124) the classification code for at least one document with the further classification code for that document;
determining (116, 125, 134) whether a disagreement exists between the assigned classification code and the further classification code for at least one identifying those documents (14a) with disagreeing classification codes as training set candidates;
applying a stop threshold (117, 138) to the training, set candidates; and grouping the training set candidates as a training set when the stop threshold is satisfied.

18. A method according to Claim 17, further comprising:
assigning further classification codes to the documents (14a) for which a disagreement exists.

19. A method according to Claim 17, wherein the classification codes are assigned by a machine and the further classification codes are assigned by a reviewer.

20. A method according to Claim 17, wherein the classification codes are assigned by a machine and the further classification codes are assigned by a further machine.