Runs - Enterprise 2005¶

baseLMlam05¶

Run ID: baseLMlam05
Participant: lowlands.devries
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Basic Language Model with lambda 0.5

baseLMlam08¶

Run ID: baseLMlam08
Participant: lowlands.devries
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Language Model with lambda 0.8.

CMUallon¶

Run ID: CMUallon
Participant: cmu.dir.callan
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Hierarchical language models using thread structure. Messages smoothed using extra weight on articles in the thread, the message it was in reply to (if applicable), the subject of the message. A prior probability was placed on the thread structure placing emphasis on messages starting the thread. Documents that were not messages but in the lists corpus were removed from the results list. Krovetz stemmer, no stopwords.

CMUnoprior¶

Run ID: CMUnoprior
Participant: cmu.dir.callan
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Hierarchical language models using thread structure. Messages smoothed using extra weight on articles in the thread, the message it was in reply to (if applicable), the subject of the message. Documents that were not messages but in the lists corpus were removed from the results list. Krovetz stemmer, no stopwords.

CMUnoPS¶

Run ID: CMUnoPS
Participant: cmu.dir.callan
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Hierarchical language models using thread structure. Messages smoothed using extra weight on articles in the thread, the subject of the message. Documents that were not messages but in the lists corpus were removed from the results list. Krovetz stemmer, no stopwords.

CMUnoPSD¶

Run ID: CMUnoPSD
Participant: cmu.dir.callan
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Hierarchical language models smoothed using extra weight on the subject of the message. Documents that were not messages but in the lists corpus were removed from the results list. Krovetz stemmer, no stopwords.

CMUnoPSDT¶

Run ID: CMUnoPSDT
Participant: cmu.dir.callan
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Simple language models. Documents that were not messages but in the lists corpus were removed from the results list. Krovetz stemmer, no stopwords.

CNDS01¶

Run ID: CNDS01
Participant: pekingu.yan
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: use pure IR technology.

CNDS02¶

Run ID: CNDS02
Participant: pekingu.yan
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: use homepage information with parameters group1

CNDS03¶

Run ID: CNDS03
Participant: pekingu.yan
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: use mail information

CNDS04LC¶

Run ID: CNDS04LC
Participant: pekingu.yan
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: combination of multiple approaches with linear method.

CNDS06¶

Run ID: CNDS06
Participant: pekingu.yan
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: use homepage information with parameters group2

CON¶

Run ID: CON
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 7/30/2005
Type: automatic
Task: discussion
Run description: The system employs the keyphrases extraction techniques as well as Natural Language Processing technqiues such as N-GRAM and Part-Of-Speech tagging. It also applies Support Vector Machine (SVM) to classify the discussion messages.

covKIRun1¶

Results | Participants | Input | Summary | Appendix

Run ID: covKIRun1
Participant: coveo.soucy
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Task: knownitem
Run description: Baseline. Queries are executed using an AND operator with Coveo Enterprise Search default settings (commercial product). If the list of results if less than 100, the list is appended using results from the same query executed with an OR operator.

covKIRun2¶

Results | Participants | Input | Summary | Appendix

Run ID: covKIRun2
Participant: coveo.soucy
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Task: knownitem
Run description: As covKIRun1, but removed emails that were replies (Title contains RE).

covKIRun3¶

Results | Participants | Input | Summary | Appendix

Run ID: covKIRun3
Participant: coveo.soucy
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Task: knownitem
Run description: As covKIRun2, but the final result list is a mix from three different queries 1st query A query with AND operator, remove replies 2th query A query with AND operator, only replies 3th query A query with the OR operator, replies or not Then, the final result list is built using a loop where 2 results from the first query are appended to the list, then one from the second, then one from the third, until the final list contains 100 results.

covKIRun4¶

Results | Participants | Input | Summary | Appendix

Run ID: covKIRun4
Participant: coveo.soucy
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Task: knownitem
Run description: As covKIRun3, but using a fourth query. This fourth query is done only when an acronym expansion is found in a list built by an acronym resolution engine provided by the NRC-ILT (National Research Council, Interactive Language Technologies). For instance, if the query is "UI language" the query is expanded as (UI OR "user interface") language. The acronyms were automatically built using the TREC text collection (with no knowledge about the train/test queries). No check was done to verify the accuracy of the acronyms.

covKIRun5¶

Results | Participants | Input | Summary | Appendix

Run ID: covKIRun5
Participant: coveo.soucy
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Task: knownitem
Run description: Like covKIRun4, but we don't remove emails that are forwards (FWD in the title). NOTE The description for covKIRun4 was incomplete. Should add the following "Also, we remove emails that are forwards (i.e contain "FWD " in the subject field)"

csiroanuds1¶

Run ID: csiroanuds1
Participant: csiro.hawking
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: BM25, no structure used

csiroanuds3¶

Run ID: csiroanuds3
Participant: csiro.hawking
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: BM25, excluding quotes/forwarded/signatures, upweight subjects

csiroanuds5¶

Run ID: csiroanuds5
Participant: csiro.hawking
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: as csiroanuds3, with disjunction of terms

csiroanuds7¶

Run ID: csiroanuds7
Participant: csiro.hawking
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: as csiroanuds1, subject line doubled

csiroanuds8¶

Run ID: csiroanuds8
Participant: csiro.hawking
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: as csiroanuds3, relevance feedback based on entire collection

csiroanuki1¶

Run ID: csiroanuki1
Participant: csiro.hawking
Track: Enterprise
Year: 2005
Submission: 8/4/2005
Task: knownitem
Run description: BM-25 based, ignoring message structure

csiroanuki2¶

Run ID: csiroanuki2
Participant: csiro.hawking
Track: Enterprise
Year: 2005
Submission: 8/4/2005
Task: knownitem
Run description: Using message structure, upweight Subject

csiroanuki3¶

Run ID: csiroanuki3
Participant: csiro.hawking
Track: Enterprise
Year: 2005
Submission: 8/4/2005
Task: knownitem
Run description: Queries transformed to include email addresses and some abbreviations

csiroanuki5¶

Run ID: csiroanuki5
Participant: csiro.hawking
Track: Enterprise
Year: 2005
Submission: 8/4/2005
Task: knownitem
Run description: Quoted material, forwarded material, and signatures ignored

csiroanuki6¶

Run ID: csiroanuki6
Participant: csiro.hawking
Track: Enterprise
Year: 2005
Submission: 8/4/2005
Task: knownitem
Run description: Text in Subject repeated

csusm1¶

Run ID: csusm1
Participant: csusm.guillen
Track: Enterprise
Year: 2005
Submission: 7/31/2005
Task: knownitem
Run description: We used the system INDRI developed by UMASS and CMU to create an index. The index was then used by "runquery" to retrieve documents. The topics were mapped to the INDRI format before running the queries.

csusm2¶

Run ID: csusm2
Participant: csusm.guillen
Track: Enterprise
Year: 2005
Submission: 7/31/2005
Task: knownitem
Run description: We used the INDRI system developed by UMASS and CMU to create an index. The index was then used to retrieve documents with the option "runquery" of INDRI. In this run we are using different combinations of ordered/unordered phrases, synonyms, single terms and exact match terms.

DrexelKI05a¶

Run ID: DrexelKI05a
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: basic tf*idf

DrexelKI05b¶

Run ID: DrexelKI05b
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: 2-pass without query expansion

DrexelKI05c¶

Run ID: DrexelKI05c
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: 2-pass with query expansion

DrexelKI05d¶

Run ID: DrexelKI05d
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: weighted query expansion

DREXEXP1¶

Run ID: DREXEXP1
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: Basically this run is based on the information retrieval Vector model. About 20,000 person names extracted from the w3c corpus as candidates for experts and each person is represented by a small collection of documents, such as emails they sent. These documents are indexed by not only keywords but also 2-gram and 3-gram phrases that are obtained from PATTree n-gram extraction algorithms. Cosine similarity is applied to measure the relevance between the queries and the candidates. During query processing, phrases in the queries were signed higher weight.

DREXEXP2¶

Run ID: DREXEXP2
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: Basically this run is based on the latent semantic indexing retrieval model. About 20,000 person names extracted from the w3c corpus as candidates for experts and each person is represented by a small collection of documents, such as emails they sent. These documents are indexed by not only keywords but also 2-gram and 3-gram phrases that are obtained from PATTree n-gram extraction algorithms. In this approach, relevance ranking is measured by matrix algebra. The technique called singular value decomposition (SVD) is applied to calculate the singular values in the large term-document matrix. The idea is that the top k singular values can be used to identify the key elements of the matrix. Each query is treated as a peudo-document. Similarity between queries and documents is scored by cosine similarity measure.

du05baseline¶

Run ID: du05baseline
Participant: uduisburg.essen.frommholz
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: In this run, no distinction is made between quoted and new text of a message. For each term t in the whole message m , the term weight is seen as the probability P(t|d) which is estimated based on tf(t,m), the term frequency of t in m, using the formula wtf(t,m) = tf(t,m)/(avgtf(m) + tf(t,m)). Retrieval is performed with probabilistic datalog in a vector-space-like manner. This run is the baseline for our other runs.

du05highstrg¶

Run ID: du05highstrg
Participant: uduisburg.essen.frommholz
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: If a message m' is a reply to message m, usually parts of m quoted by the author of m'. These quotations are interpreted as highlightings of parts of m (further referred to as "highlight quotes"). For a message m, each quotation q of a successor message m' is extracted and linked with m. Each q is seen as a separate document; for each document d (containing m and q), term weights are the estimated probability P(t|d), using tf(t,d), the term frequency of t in d and the formula P(t|d) \approx wtf(t,d) = tf(t,d)/(avgtf(d) + tf(t,d)). Another component is the link between m and q, which can be weak or strong, reflecting the influence the highlight quotation has on m's term weights. The link weight is expressed by P(link(q,m)). For a message m, the overall term weights P_{high}(t|m) are thus estimated as P(t|m) + \sum_q P(link(q,m)) * P(t|q). In this run, it is P(link(q,m)) = 0.9, so we have a strong connection between m and each q. Retrieval is performed with probabilistic datalog in a vector-space-like manner.

du05highweak¶

Run ID: du05highweak
Participant: uduisburg.essen.frommholz
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: If a message m' is a reply to message m, usually parts of m quoted by the author of m'. These quotations are interpreted as highlightings of parts of m (further referred to as "highlight quotes"). For a message m, each quotation q of a successor message m' is extracted and linked with m. Each q is seen as a separate document; for each document d (containing m and q), term weights are the estimated probability P(t|d), using tf(t,d), the term frequency of t in d and the formula P(t|d) \approx wtf(t,d) = tf(t,d)/(avgtf(d) + tf(t,d)). Another component is the link between m and q, which can be weak or strong, reflecting the influence the highlight quotation has on m's term weights. The link weight is expressed by P(link(q,m)). For a message m, the overall term weights P_{high}(t|m) are thus estimated as P(t|m) + \sum_q P(link(q,m)) * P(t|q). In this run, it is P(link(q,m)) = 0.3, so we only have a weak connection between m and each q. Retrieval is performed with probabilistic datalog in a vector-space-like manner.

du05quotstrg¶

Run ID: du05quotstrg
Participant: uduisburg.essen.frommholz
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: The goal of this run is to study the effect of the quotation part of the message on retrieval effectiveness. During indexing, an email message m is separated into two parts the quotation quot and the message mes (containing only the new parts of the message). quot and mes are seen as separate documents which are connected via a link. For each term, the term weight is seen as the probability P(t|d) which is estimated based on tf(t,quot) and tf(t,mes), the term frequency of t in quot and in mes, respectively, using the formula wtf(t,d) = tf(t,d)/(avgtf(d) + tf(t,d)). Another component is the link between quot and mes, which can be weak or strong, reflecting the influence the quotation part has on term weights. The link weight is expressed by P(link(quot,mes)). For a message m, P(t|m) is thus estimated as P(t|mes) + P(link(quot,mess)) * P(t|quot). In this run, it is P(link(quot,mess)) = 0.9, so we have a strong connection between mes and quot. Retrieval is performed with probabilistic datalog in a vector-space-like manner.

du05quotweak¶

Run ID: du05quotweak
Participant: uduisburg.essen.frommholz
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: The goal of this run is to study the effect of the quotation part of the message on retrieval effectiveness. During indexing, an email message m is separated into two parts the quotation quot and the message mes (containing only the new parts of the message). quot and mes are seen as separate documents which are connected via a link. For each term, the term weight is seen as the probability P(t|d) which is estimated based on tf(t,quot) and tf(t,mes), the term frequency of t in quot and in mes, respectively, using the formula wtf(t,d) = tf(t,d)/(avgtf(d) + tf(t,d)). Another component is the link between quot and mes, which can be weak or strong, reflecting the influence the quotation part has on term weights. The link weight is expressed by P(link(quot,mes)). For a message m, P(t|m) is thus estimated as P(t|mes) + P(link(quot,mess)) * P(t|quot). In this run, it is P(link(quot,mess)) = 0.3, so we only have a weak connection between mes and quot. Retrieval is performed with probabilistic datalog in a vector-space-like manner.

humEK05l¶

Results | Participants | Input | Summary | Appendix

Run ID: humEK05l
Participant: hummingbird.tomlinson
Track: Enterprise
Year: 2005
Submission: 8/1/2005
Task: knownitem
Run description: plain content search including linguistic expansion from English inflectional stemming

humEK05p¶

Results | Participants | Input | Summary | Appendix

Run ID: humEK05p
Participant: hummingbird.tomlinson
Track: Enterprise
Year: 2005
Submission: 8/1/2005
Task: knownitem
Run description: same as humEK05pl except inflections disabled

humEK05pl¶

Results | Participants | Input | Summary | Appendix

Run ID: humEK05pl
Participant: hummingbird.tomlinson
Track: Enterprise
Year: 2005
Submission: 8/1/2005
Task: knownitem
Run description: same as humEK05tl except additional meta fields included in addition to Title (similar to last year's humW04pl)

humEK05t3l¶

Results | Participants | Input | Summary | Appendix

Run ID: humEK05t3l
Participant: hummingbird.tomlinson
Track: Enterprise
Year: 2005
Submission: 8/1/2005
Task: knownitem
Run description: same as humEK05tl except 3x extra weight on Title

humEK05tl¶

Results | Participants | Input | Summary | Appendix

Run ID: humEK05tl
Participant: hummingbird.tomlinson
Track: Enterprise
Year: 2005
Submission: 8/1/2005
Task: knownitem
Run description: same as humEK05l except extra weight on Title

irmdLT¶

Results | Participants | Input | Summary | Appendix

Run ID: irmdLT
Participant: umagdeburg.nuernberger
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: For retrieval the language modeling approach was used (two-stage LM). The top results of an inital content only run were used to retrieve all documents that appeared in the same threads as the top retrieved ones and reranking was performed on this set.

irmdLTF¶

Results | Participants | Input | Summary | Appendix

Run ID: irmdLTF
Participant: umagdeburg.nuernberger
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: For retrieval the language modeling approach was used (two-stage LM). The top results of an inital content only run were used to retrieve all documents that appeared in the same threads as the top retrieved ones and reranking was performed on this set. Also, documents were filtered based on the characteristics of the document's term frequency distribution.

irmdT¶

Results | Participants | Input | Summary | Appendix

Run ID: irmdT
Participant: umagdeburg.nuernberger
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: For retrieval the language modeling approach was used (two-stage LM).

irmdTTF¶

Results | Participants | Input | Summary | Appendix

Run ID: irmdTTF
Participant: umagdeburg.nuernberger
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: For retrieval the language modeling approach was used (two-stage LM). The top results of an inital content only run were used to retrieve all documents that appeared in the same threads as the top retrieved ones and reranking was performed on this set. Also, documents were filtered based on the characteristics of the document's term frequency distribution.

KIDEFAULT¶

Run ID: KIDEFAULT
Participant: umaryland.oard
Track: Enterprise
Year: 2005
Submission: 8/1/2005
Task: knownitem
Run description: Default baseline run with INQUERY.

KITHRNOQUOT¶

Run ID: KITHRNOQUOT
Participant: umaryland.oard
Track: Enterprise
Year: 2005
Submission: 8/1/2005
Task: knownitem
Run description: Document expansion with threads without quoted text. If a message is in a thread which is composed of at least two messages, the message is expanded with one copy of itself, one copy of its chronologically previous message, and one copy of its chronologically next message. Quoted texts are removed from the messages.

KITHRQUOT¶

Run ID: KITHRQUOT
Participant: umaryland.oard
Track: Enterprise
Year: 2005
Submission: 8/1/2005
Task: knownitem
Run description: Document expansion with threads with quoted text. If a message is in a thread which is composed of at least two messages, the message is expanded with one copy of itself, one copy of its chronologically previous message, and one copy of its chronologically next message. Quoted texts are kept in the messages.

KITRANS¶

Run ID: KITRANS
Participant: umaryland.oard
Track: Enterprise
Year: 2005
Submission: 8/1/2005
Task: knownitem
Run description: Date and version expressions in both the documents and the queries have been normalized. For instance, 12-Mar-2000 is transformed to day12 March 2000; HTML 4.0 is transformed to HTML four pointx zero.

KNOWNITEM¶

Run ID: KNOWNITEM
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: We applied Support Vector Machines to known item task. We also applied N-Gram and POS tagging and other Natural Language Processing techniques.

LLEXemails¶

Run ID: LLEXemails
Participant: lowlands.devries
Track: Enterprise
Year: 2005
Submission: 10/4/2005
Task: expert
Run description: Baseline run, based on Language models estimated from emails sent by candidates.

LMheaderAND¶

Run ID: LMheaderAND
Participant: lowlands.devries
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Language Model with lambda 0.85. Combination of 'header' and 'text' scores by 'and' operator (product).

LMheaderOR¶

Run ID: LMheaderOR
Participant: lowlands.devries
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Language Model with lambda 0.85. Combination of 'header' and 'text' scores by 'or' operator (sum).

LMlam08Thr¶

Run ID: LMlam08Thr
Participant: lowlands.devries
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Language Model with lambda 0.8. Propagation of threads scores.

LMMlam05Thr¶

Run ID: LMMlam05Thr
Participant: lowlands.devries
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Language Model with lambda 0.5. Propagation of threads scores.

LMplaintext¶

Run ID: LMplaintext
Participant: lowlands.devries
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Language Model with lambda 0.85. Simple Document ranking.

LMsubjectAND¶

Run ID: LMsubjectAND
Participant: lowlands.devries
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Language Model with lambda 0.85. Combination of 'subject' and 'text' scores by 'and' operator (product).

LMsubjectOR¶

Run ID: LMsubjectOR
Participant: lowlands.devries
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Language Model with lambda 0.85. Combination of 'subject' and 'text' scores by 'or' operator (sum).

LPC1¶

Results | Participants | Input | Summary | Appendix

Run ID: LPC1
Participant: langpower.yang
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: manual
Task: discussion
Run description: Briefly list the most salient features of this run.

LPC2¶

Results | Participants | Input | Summary | Appendix

Run ID: LPC2
Participant: langpower.yang
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: manual
Task: discussion
Run description: Briefly list the most salient features of this run.

LPC3¶

Results | Participants | Input | Summary | Appendix

Run ID: LPC3
Participant: langpower.yang
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: System built using rule-based NLP and concept-based indexing. Earlier processing with lower coverage in parsing and indexing.

LPC4¶

Results | Participants | Input | Summary | Appendix

Run ID: LPC4
Participant: langpower.yang
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: System built using rule-based NLP and concept-based indexing.

LPC5¶

Results | Participants | Input | Summary | Appendix

Run ID: LPC5
Participant: langpower.yang
Track: Enterprise
Year: 2005
Submission: 8/6/2005
Task: knownitem
Run description: Briefly list the most salient features of this run.

MSRA051¶

Run ID: MSRA051
Participant: microsoft.cao
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: The basic model, 1) a two-stage model of combining relevance and co-occurrence 2) the co-occurrence model consists of body-body, title-author, and title-tree submodels 3) a back-off query term matching method which prefers exact match, then partial match, and finally word-level match

MSRA052¶

Run ID: MSRA052
Participant: microsoft.cao
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: The basic model plus acronym normalization

MSRA053¶

Run ID: MSRA053
Participant: microsoft.cao
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: The basic model without acronym indexing

MSRA054¶

Run ID: MSRA054
Participant: microsoft.cao
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: The basic model plus cluster-based re-ranking

MSRA055¶

Run ID: MSRA055
Participant: microsoft.cao
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: The basic model plus acronym normalization and cluster-based re-ranking

MSRCDS1¶

Run ID: MSRCDS1
Participant: microsoft.robertson
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Three text fields Subject, Body, QuotedBody. Two static features HasParent and Year.

MSRCDS2¶

Run ID: MSRCDS2
Participant: microsoft.robertson
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Three text fields Subject, Body, QuotedBody.

MSRCDS3¶

Run ID: MSRCDS3
Participant: microsoft.robertson
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Three text fields but with even weights Subject, Body, QuotedBody.

MSRCDS4¶

Run ID: MSRCDS4
Participant: microsoft.robertson
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Three text fields Subject, Body, QuotedBody. Three static features HasParent, Year, URLsInBody and SubjectContainsRE.

MSRCDS5¶

Run ID: MSRCDS5
Participant: microsoft.robertson
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Three text fields Subject+From, Body, QuotedBody. Four static features HasParent, Year, URLsInBody and SubjectContainsRE.

MSRCKI1¶

Run ID: MSRCKI1
Participant: microsoft.robertson
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Task: knownitem
Run description: Three text fields Subject, Body, QuotedBody. Two static features HasParent and Year.

MSRCKI2¶

Run ID: MSRCKI2
Participant: microsoft.robertson
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Task: knownitem
Run description: Three text fields Subject, Body, QuotedBody.

MSRCKI3¶

Run ID: MSRCKI3
Participant: microsoft.robertson
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Task: knownitem
Run description: Three text fields but with even weights Subject, Body, QuotedBody.

MSRCKI4¶

Run ID: MSRCKI4
Participant: microsoft.robertson
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Task: knownitem
Run description: Three text fields Subject, Body, QuotedBody. Four static features HasParent, Year, URLsInBody and SubjectContainsRE.

MSRCKI5¶

Run ID: MSRCKI5
Participant: microsoft.robertson
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Task: knownitem
Run description: Three text fields Subject+From, Body, QuotedBody. Four static features HasParent, Year, URLsInBody and SubjectContainsRE.

MU05ENd1¶

Run ID: MU05ENd1
Participant: umelbourne.anh
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Dirichlet metric. Document scores enhanced with decaying-weighted scores of thread ancestors, descendants.

MU05ENd2¶

Run ID: MU05ENd2
Participant: umelbourne.anh
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Dirichlet metric. Body of message and quoted text held in parallel indexes, score of document a weighted combination of the two.

MU05ENd3¶

Run ID: MU05ENd3
Participant: umelbourne.anh
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Dirichlet metric. Message score enhanced by "authority" of author; authority of author judged by number of postings by that author.

MU05ENd4¶

Run ID: MU05ENd4
Participant: umelbourne.anh
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Dirichlet metric on body (not quoted text) of messages.

MU05ENd5¶

Run ID: MU05ENd5
Participant: umelbourne.anh
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Impact metric on body of messages (not quoted text).

NON¶

Run ID: NON
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 7/30/2005
Type: automatic
Task: discussion
Run description: The system employs the keyphrases extraction techniques as well as Natural Language Processing technqiues such as N-GRAM and Part-Of-Speech tagging. It also applies Support Vector Machine (SVM) to classify the discussion messages.

OddsC¶

Run ID: OddsC
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Infers structural components of query and ranks using the Odds ratio of each component. Automatic Run.

OddsWcEst¶

Run ID: OddsWcEst
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Infers structural components of query and ranks using the Odds ratio of each component. Uses weighted components with are automatically estimated. Automatic Run.

PITTDTA1¶

Run ID: PITTDTA1
Participant: upittsburgh.he
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: performed document expansion using w3c-www-esw-people collection. used Indri2.0 search engine

PITTDTA2BIG1¶

Run ID: PITTDTA2BIG1
Participant: upittsburgh.he
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: performed document expansion using w3c-www-esw-people collection. used Indri2.0 search engine, also performed document reranking using the thread information (big reranking algorithm).

PITTDTA2SML1¶

Run ID: PITTDTA2SML1
Participant: upittsburgh.he
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: performed document expansion using w3c-www-esw-people collection. used Indri2.0 search engine, also performed document reranking using the thread information (small reranking algorithm)

PITTDTA2SML2¶

Run ID: PITTDTA2SML2
Participant: upittsburgh.he
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: performed document expansion using w3c-www-esw-people collection. used Indri2.0 search engine, also performed document reranking using the thread information (small reranking algorithm), reranking used top 50 docs with adjust weight 0.8

PITTKIA1W7¶

Run ID: PITTKIA1W7
Participant: upittsburgh.he
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: use indri2.0 as the search engine, use w3c-www-esw-people collection as the document expansion collection for generating top 20 terms from top 20 documents. then search the w3c-email collection. the weight between the original query and the expanded part is 0.7 and 0.3

PITTKIA1W8¶

Run ID: PITTKIA1W8
Participant: upittsburgh.he
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: use indri2.0 as the search engine, use w3c-www-esw-people collection as the document expansion collection for generating top 20 terms from top 20 documents. then search the w3c-email collection. the weight between the original query and the expanded part is 0.8 and 0.2

PITTKINB1¶

Run ID: PITTKINB1
Participant: upittsburgh.he
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: use indri2.0 as the search engine, search the w3c-email collection directly.

PITTKIWB2¶

Run ID: PITTKIWB2
Participant: upittsburgh.he
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: use indri2.0 as the search engine, search the w3c-email collection directly. use Indri's blind relevance feedback feature with the parameter as look at top 20 documents, select top 20 terms, with weight 0.5

prisds1¶

Run ID: prisds1
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: A result for discussion task produced by group of Pattern Recognition & Information System, based lemur.

prisds2¶

Run ID: prisds2
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: A result for discussion task produced by group of Pattern Recognition & Information System, based lemur.

prisds3¶

Run ID: prisds3
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: A result for discussion task produced by group of Pattern Recognition & Information System, based lemur.

prisds4¶

Run ID: prisds4
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: A result for discussion task produced by group of Pattern Recognition & Information System, based lemur.

prisds5¶

Run ID: prisds5
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 7/28/2005
Type: automatic
Task: discussion
Run description: A result for discussion task produced by group of Pattern Recognition & Information System, based lemur.

PRISEX1¶

Run ID: PRISEX1
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 10/10/2005
Task: expert
Run description: BASED ON LEMUR3.1

PRISEX2¶

Run ID: PRISEX2
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 10/10/2005
Task: expert
Run description: BASED ON LEMUR3.1

PRISEX3¶

Run ID: PRISEX3
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 10/10/2005
Task: expert
Run description: BASED ON LEMUR3.1

PRISEX4¶

Run ID: PRISEX4
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 10/10/2005
Task: expert
Run description: BASED ON LEMUR3.1

PRISEX5¶

Run ID: PRISEX5
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 10/10/2005
Task: expert
Run description: BASED ON LEMUR3.1

priski1¶

Run ID: priski1
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 8/2/2005
Task: knownitem
Run description: A result for the known item task produced by the group of Pattern Recognition & Information System, based lemur.(The number is not the order.)

priski2¶

Run ID: priski2
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 8/2/2005
Task: knownitem
Run description: A result for the known item task produced by the group of Pattern Recognition & Information System, based lemur.(The number is not the order.)

priski3¶

Run ID: priski3
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 8/2/2005
Task: knownitem
Run description: A result for the known item task produced by the group of Pattern Recognition & Information System, based lemur.(The number is not the order.)

priski4¶

Run ID: priski4
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 8/2/2005
Task: knownitem
Run description: A result for the known item task produced by the group of Pattern Recognition & Information System, based lemur.(The number is not the order.)

priski5¶

Run ID: priski5
Participant: beijingu.guo
Track: Enterprise
Year: 2005
Submission: 8/2/2005
Task: knownitem
Run description: A result for the known item task produced by the group of Pattern Recognition & Information System, based lemur.(The number is not the order.)

PRO¶

Run ID: PRO
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 7/30/2005
Type: automatic
Task: discussion
Run description: The system employs the keyphrases extraction techniques as well as Natural Language Processing technqiues such as N-GRAM and Part-Of-Speech tagging. It also applies Support Vector Machine (SVM) to classify the discussion messages.

qdC¶

Run ID: qdC
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Infers structural components of query and ranks using the likelihood of each component. Automatic Run.

qdFlat¶

Run ID: qdFlat
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Baseline run Using Language Modelling approach with Jelinek Mercer Smoothing where lambda = 0.1. All email fields indexed and porter stemming applied to corpus. Porter stemming and stopping applied to queries. Fully automatic run.

qdWcEst¶

Run ID: qdWcEst
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Infers structural components of query and ranks using the likelihood of each component. Uses weighted components with are automatically estimated. Automatic Run.

qmirdju¶

Run ID: qmirdju
Participant: qm-univ-london.roelleke
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: tf-idf based strategy, with disjoint(normalized query term weights) assumption for the probabilistic combination of idf based probabilities, unstemmed

qmirdts¶

Run ID: qmirdts
Participant: qm-univ-london.roelleke
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: tf-idf based strategy, with independence assumption for the probabilistic combination of idf based probabilities, stemmed

qmirdtu¶

Run ID: qmirdtu
Participant: qm-univ-london.roelleke
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: tf-idf based strategy, with independence assumption for the probabilistic combination of idf based probabilities, unstemmed

qmirex1¶

Run ID: qmirex1
Participant: qm-univ-london.roelleke
Track: Enterprise
Year: 2005
Submission: 10/4/2005
Task: expert
Run description: The retrieval strategy has two steps, the first step retrieve relevant emails, and the second step find the experts from the top 1000 relevant emails. We describe LCF (Local Contribution Frequency) and GCF (Global Contribution Frequency) to be used to rank the experts.

qmirex2¶

Run ID: qmirex2
Participant: qm-univ-london.roelleke
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: The strategy first finds the relevant documents, then the person who wrote the most relevant emails will be ranked highest.

qmirex3¶

Run ID: qmirex3
Participant: qm-univ-london.roelleke
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: The strategy first retrieves relevant documents, we use similar strategy with qmirex1, but here we find the authors who wrote the most relevant emails to cumpute the lcf rather than compute the authors' average weight.

qmirex4¶

Run ID: qmirex4
Participant: qm-univ-london.roelleke
Track: Enterprise
Year: 2005
Submission: 10/6/2005
Task: expert
Run description: lcf-gcf strategy with amended implementation

qmirkidju¶

Run ID: qmirkidju
Participant: qm-univ-london.roelleke
Track: Enterprise
Year: 2005
Submission: 8/4/2005
Task: knownitem
Run description: tf-idf based strategy, with disjoint(normalized query term weights) assumption for the probabilistic combination of idf based probabilities, unstemmed

qmirkidtu¶

Run ID: qmirkidtu
Participant: qm-univ-london.roelleke
Track: Enterprise
Year: 2005
Submission: 8/4/2005
Task: knownitem
Run description: tf-idf based strategy, with independence assumption for the probabilisitic combination of idf based probabilities, unstemmed

THRNOQNARR¶

Run ID: THRNOQNARR
Participant: umaryland.oard
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Document expansion with threads without quoted text. If a message is in a thread which is composed of at least two messages, the message is expanded with one copy of itself, one copy of its chronologically previous message, and one copy of its chronologically next message. Quoted texts are removed from the messages. In this run, queries are taken from both the "query" and "narrative" fields.

THRNOQUOT¶

Run ID: THRNOQUOT
Participant: umaryland.oard
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Document expansion with threads without quoted text. If a message is in a thread which is composed of at least two messages, the message is expanded with one copy of itself, one copy of its chronologically previous message, and one copy of its chronologically next message. Quoted texts are removed from the messages. In this run, queries are taken from both the "query" fields only.

THRQUOT¶

Run ID: THRQUOT
Participant: umaryland.oard
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Document expansion with threads with quoted text. If a message is in a thread which is composed of at least two messages, the message is expanded with one copy of itself, one copy of its chronologically previous message, and one copy of its chronologically next message. Quoted texts are kept in the messages.

THUENT0501¶

Run ID: THUENT0501
Participant: tsinghua.ma
Track: Enterprise
Year: 2005
Submission: 10/3/2005
Task: expert
Run description: This run makes use of all w3c web part information and inlink anchor text of these files. Text content are reconstructed and formed description files for each candidate person. Structure information inside web pages was also used to improve performance

THUENT0502¶

Run ID: THUENT0502
Participant: tsinghua.ma
Track: Enterprise
Year: 2005
Submission: 10/3/2005
Task: expert
Run description: This run makes use of all w3c web part information and inlink anchor text of these files. Text content are reconstructed and formed description files for each candidate person. Structure information inside web pages was also used to improve performance. Words from important pages are emphasized in this run.

THUENT0503¶

Run ID: THUENT0503
Participant: tsinghua.ma
Track: Enterprise
Year: 2005
Submission: 10/3/2005
Task: expert
Run description: This run makes use of all w3c web part information and inlink anchor text of these files. Text content are reconstructed and formed description files for each candidate person. Structure information inside web pages was also used to improve performance. Words from important pages are emphasized in this run. Word pairs are also given a higher weight.

THUENT0504¶

Run ID: THUENT0504
Participant: tsinghua.ma
Track: Enterprise
Year: 2005
Submission: 10/3/2005
Task: expert
Run description: This run makes use of all w3c web part information and Email lists (the list part) together with inlink anchor text of these files. Text content are reconstructed and formed description files for each candidate person. Structure information inside web pages was also used to improve performance. Words from important pages are emphasized in this run.

THUENT0505¶

Run ID: THUENT0505
Participant: tsinghua.ma
Track: Enterprise
Year: 2005
Submission: 10/3/2005
Task: expert
Run description: This run makes use of all w3c web part information and Email lists (the list part) together with inlink anchor text of these files. Text content are reconstructed and formed description files for each candidate person. Structure information inside web pages was also used to improve performance. Words from important pages are emphasized in this run. Bi-gram retrieval was also applied.

TITLEDEFAULT¶

Run ID: TITLEDEFAULT
Participant: umaryland.oard
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Default baseline run with INQUERY.

TITLETRANS¶

Run ID: TITLETRANS
Participant: umaryland.oard
Track: Enterprise
Year: 2005
Submission: 7/30/2005
Type: automatic
Task: discussion
Run description: Date and version expressions in both the documents and the queries have been normalized. For instance, 12-Mar-2000 is transformed to day12 March 2000; HTML 4.0 is transformed to HTML four pointx zero.

ToNsBs350¶

Run ID: ToNsBs350
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Query Field only (with stopping) No Stemming applied Language Modelling approach using Bayes Smoothing

ToNsBs350F¶

Run ID: ToNsBs350F
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: From the baseline run, we filter out unimportant messages

ToNsBs350FT¶

Run ID: ToNsBs350FT
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Baseline + Filter and prior according to thread size

ToNsBs350FT5¶

Run ID: ToNsBs350FT5
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: Baseline + Filter and prior according to thread size

uams05run0¶

Run ID: uams05run0
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 10/3/2005
Task: expert
Run description: Considering the top 500 documents for each topic. Build a language model for each query for each candidate.

uams05run1¶

Run ID: uams05run1
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: S(1) top200documents,EXACT_MATCH,dirichletLM,beta=50000

uams05run2¶

Run ID: uams05run2
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: S(2) top200documents,EMAIL_MATCH,dirichletLM,beta=50000

uams05run3¶

Run ID: uams05run3
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: S(3) top200documents,EXACT_MATCH0.5 + EMAIL_MATCH0.5,dirichletLM,beta=50000

uams05run4¶

Run ID: uams05run4
Participant: uamsterdam.mueller
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: S(4) top200documents,EXACT_MATCH0.375 + NAME_MATCH0.208 + EMAIL_MATCH*0.416,dirichletLM,beta=50000

UNKNOWN¶

Run ID: UNKNOWN
Participant: drexelu.allen
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: We applied Support Vector Machines to known item task. We also applied N-Gram and POS tagging and other Natural Language Processing techniques.

uogEBase¶

Run ID: uogEBase
Participant: uglasgow.ounis
Track: Enterprise
Year: 2005
Submission: 7/30/2005
Task: knownitem
Run description: Divergence From Randomness PL2F weighting model with field specific term-frequency normaliastion, 3 fields (body, title, anchor text). Index only lists sub-collection. Weak porter stemming.

uogEDates1¶

Run ID: uogEDates1
Participant: uglasgow.ounis
Track: Enterprise
Year: 2005
Submission: 7/30/2005
Task: knownitem
Run description: Divergence From Randomness PL2F weighting model with field specific term-frequency normaliastion, 3 fields (body, title, anchor text). Index only lists sub-collection. Weak porter stemming. Boosting messages that occur near to dates mentioned in the topics.

uogEDates12T¶

Run ID: uogEDates12T
Participant: uglasgow.ounis
Track: Enterprise
Year: 2005
Submission: 7/30/2005
Task: knownitem
Run description: Divergence From Randomness PL2F weighting model with field specific term-frequency normaliastion, 3 fields (body, title, anchor text). Index only lists sub-collection. Weak porter stemming. Boosting more recent messages, messages that were sent near to dates mentioned in the topics, and messages relative to their thread depth.

uogEDates2¶

Run ID: uogEDates2
Participant: uglasgow.ounis
Track: Enterprise
Year: 2005
Submission: 7/30/2005
Task: knownitem
Run description: Divergence From Randomness PL2F weighting model with field specific term-frequency normaliastion, 3 fields (body, title, anchor text). Index only lists sub-collection. Weak porter stemming. Boosting more recent messages.

uogES05B2¶

Run ID: uogES05B2
Participant: uglasgow.ounis
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: Combining documents with 2 normalisations of term frequencies. Based on Divergence from Randomness framework.

uogES05CbaDT¶

Run ID: uogES05CbaDT
Participant: uglasgow.ounis
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: Combining documents with multiple normalisations, three content fields. Based on Divergence from Randomness framework. Uses additional email evidence - threads and dates.

uogES05Cbase¶

Run ID: uogES05Cbase
Participant: uglasgow.ounis
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: Combining documents with multiple normalisations, three content fields. Based on Divergence from Randomness framework.

uogES05CbiH¶

Run ID: uogES05CbiH
Participant: uglasgow.ounis
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: Combining documents with multiple normalisations, three content fields. Based on Divergence from Randomness framework. Uses additional web evidence - homepages.

uogES05Cbis¶

Run ID: uogES05Cbis
Participant: uglasgow.ounis
Track: Enterprise
Year: 2005
Submission: 10/5/2005
Task: expert
Run description: Combining documents with multiple normalisations, three content fields. Based on Divergence from Randomness framework.

UwatEntDS¶

Run ID: UwatEntDS
Participant: uwaterloo.vechtomova
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: - Okapi was used for retrieval; - Only original text in the messages was indexed.

UwatEntDSq¶

Run ID: UwatEntDSq
Participant: uwaterloo.vechtomova
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: - Okapi was used for retrieval; - Complete messages were indexed (with quoted text).

UwatEntDSth¶

Run ID: UwatEntDSth
Participant: uwaterloo.vechtomova
Track: Enterprise
Year: 2005
Submission: 7/29/2005
Type: automatic
Task: discussion
Run description: - Okapi was used for retrieval; - Complete threads were indexed; - Only original text in the messages was indexed.

UWATEntKI¶

Run ID: UWATEntKI
Participant: uwaterloo.vechtomova
Track: Enterprise
Year: 2005
Submission: 8/5/2005
Task: knownitem
Run description: Documents ranked using Okapi.

WIMent01¶

Run ID: WIMent01
Participant: fudan.niu
Track: Enterprise
Year: 2005
Submission: 8/2/2005
Task: knownitem
Run description: KL-divergence language model based retrieval method, using author, date, to do rerank