Runs - Novelty 2004¶

ccs1f0t1¶

Run ID: ccs1f0t1
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: HMM with 9 states and 1 features, log(#signature terms+1) Threshold 0 for novel Did not use topic description at all for this run.

ccs1ftop0t1¶

Run ID: ccs1ftop0t1
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: HMM with 9 states and 1 features, log(#signature terms+1) choose top 25 docs based on sum gamma probabilities Threshold 0 for novel (no QR) Did not use topic description at all for this run.

ccs3fbqrt3¶

Run ID: ccs3fbqrt3
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: no topic description used Used first 5 docs generate new signature terms build HMM and then selected sentence used QR to find novel

ccs3fmmr95t3¶

Run ID: ccs3fmmr95t3
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: no topic description used HMM with 3 features, followed by MMR

ccs3fmmrt1¶

Run ID: ccs3fmmrt1
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: HMM with 9 states and 3 features, log(w+1), sentence entropy, log(#signature terms+1) choose top 25 docs based on sum gamma probabilities MMR for Novel sentence Did not use topic description at all for this run.

ccs3fmmrt3¶

Run ID: ccs3fmmrt3
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: no topic description used Used first 5 docs to build HMM and then selected sentence used MMR to find novel

ccs3fqrt1¶

Run ID: ccs3fqrt1
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: HMM with 9 states and 3 features. No topic information used.

ccs3ftop0t1¶

Run ID: ccs3ftop0t1
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: HMM with 9 states and 3 features, log(w+1), sentence entropy, log(#signature terms+1) choose top 25 docs based on sum gamma probabilities Threshold 0 for novel (no QR) Did not use topic description at all for this run.

ccsfbmmrt3¶

Run ID: ccsfbmmrt3
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: no topic description used Used first 5 docs to extend signature terms rebuilt HMM and then selected sentence used MMR to find novel

ccsmmr2t2¶

Run ID: ccsmmr2t2
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: no topic description used MMR

ccsmmr3t2¶

Run ID: ccsmmr3t2
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: no topic description used MMR lambda=0.3

ccsmmr4t2¶

Run ID: ccsmmr4t2
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: no topic description used MMR lambda=0.4

ccsmmr5t2¶

Run ID: ccsmmr5t2
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: no topic description used MMR lambda=0.5

ccsqrt2¶

Run ID: ccsqrt2
Participant: ida.ccs.nsa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: no topic description used pivoted QR

cdvp4CnQry2¶

Run ID: cdvp4CnQry2
Participant: dubblincity.u
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: BaseLine experiment no Qe no Doce Unique terms identification

cdvp4CnS101¶

Run ID: cdvp4CnS101
Participant: dubblincity.u
Track: Novelty
Year: 2004
Submission: 9/2/2004
Task: task1
Run description: Uses concataned queries over Similarilty Doc Expansion getting unique words and clearing the history.

cdvp4NSen4¶

Run ID: cdvp4NSen4
Participant: dubblincity.u
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: Alogrithm called NewSentenceValue using the idfs of the unique terms together with idf of previously seen terms in the history.

cdvp4NSnoH4¶

Run ID: cdvp4NSnoH4
Participant: dubblincity.u
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: Algorithm NewSenteneceValue comparing pairwise arocross all sentences encountered.

cdvp4NTerFr1¶

Run ID: cdvp4NTerFr1
Participant: dubblincity.u
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: Alogrithm called NewnessTermFreq using the tfs of the unique terms together with idf of previously seen terms in the history.

cdvp4NTerFr3¶

Run ID: cdvp4NTerFr3
Participant: dubblincity.u
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: Alogrithm called NewnTermFreq using the tfs of the unique terms together with idf of previously seen terms in the history.

cdvp4QePDPC2¶

Run ID: cdvp4QePDPC2
Participant: dubblincity.u
Track: Novelty
Year: 2004
Submission: 9/2/2004
Task: task1
Run description: Qe using Proximity Docexp using Proximity. The Query is Concatenated Unique terms identification

cdvp4QePnD2¶

Run ID: cdvp4QePnD2
Participant: dubblincity.u
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: Query Expansion using Proximity 10 words. Unique terms identification

cdvp4QeSnD1¶

Run ID: cdvp4QeSnD1
Participant: dubblincity.u
Track: Novelty
Year: 2004
Submission: 9/2/2004
Task: task1
Run description: Query expansion (Similarity of words) over a doc using newSentence variable to determine threshold of novel sentences.

cdvp4UnHis3¶

Run ID: cdvp4UnHis3
Participant: dubblincity.u
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: Baseline approach- overlapping of word

CIIRT1R1¶

Run ID: CIIRT1R1
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: TFIDF techniques with query expansion were used for finding relevant sentences and new words and named entities were considered for detecting novel sentences.

CIIRT1R2¶

Run ID: CIIRT1R2
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: TFIDF techniques with selective query expansion were used for finding relevant sentences and new words and named entities were considered for detecting novel sentences.

CIIRT1R3¶

Run ID: CIIRT1R3
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: TFIDF techniques with query expansion and short sentences removal were used for finding relevant sentences, and specific named entities were considered for identifying redundant sentences.

CIIRT1R5¶

Run ID: CIIRT1R5
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: Take best sentence for each topic and pick top 25 best sentences. The 25 corresponding documents were chosen for further relevant sentences retrieval and novel sentences detection.

CIIRT1R6¶

Run ID: CIIRT1R6
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: For each topic, 25 documents were chosen based on the ranking of the best sentence in each document and selective query expansion was performed for relevant sentences retrieval.

CIIRT2R1¶

Run ID: CIIRT2R1
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: Pairwise sentence similarity was used for identifying novel sentences.

CIIRT2R2¶

Run ID: CIIRT2R2
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: Pairwise sentence similarity was used for identifying novel sentences with the threshold of 0.6.

CIIRT3R1¶

Run ID: CIIRT3R1
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/13/2004
Task: task3
Run description: TFIDF techniques with selective relevance feedback were used for finding relevant sentences, new words and named entities were considered for identifying novel sentences.

CIIRT3R2¶

Run ID: CIIRT3R2
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task3
Run description: TFIDF techniques with relevance feedback were used for finding relevant sentences, new words and named entities were considered for identifying novel sentences.

CIIRT3R3¶

Run ID: CIIRT3R3
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task3
Run description: TFIDF techniches with selective relevance feedback were used for finding relevant sentences and sentence pairwise similarity was consider for identify novel sentences.

CIIRT3R4¶

Run ID: CIIRT3R4
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task3
Run description: TFIDF techniques with selective relevance feedback were used for finding relevant sentences,pairwise sentence similarity was used for identifying novel sentences with the threshold of 0.6.

CIIRT3R5¶

Run ID: CIIRT3R5
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: This approach tunes similarity threshold for each topic based on its performance within the given top 5 documents for identifying novel sentences in the remaining documents.

CIIRT4R1¶

Run ID: CIIRT4R1
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/20/2004
Task: task4
Run description: Sentence pairwise similarity was used for novelty detection and different cutoff threshold was chosen for each topic accoroding its performance within the top 5 given documents.

CIIRT4R2¶

Run ID: CIIRT4R2
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/21/2004
Task: task4
Run description: Both sentence pariwise similarity and overlap of named entities were considered for identifying novel sentences.

CIIRT4R3¶

Run ID: CIIRT4R3
Participant: u.mass
Track: Novelty
Year: 2004
Submission: 9/22/2004
Task: task4
Run description: Sentence pairwise similarity and percentage of named entities overlap were considered for identifying novel sentneces, parameters were tuned with the given judgment of the top 5 documents for each topic.

clr04n1h2¶

Run ID: clr04n1h2
Participant: clresearch
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: Topics and documents were fully parsed and processed into an XML representation; each sentence was evaluated against the topic using string, syntax, and semantic characteristics.

clr04n1h3¶

Run ID: clr04n1h3
Participant: clresearch
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: Topics and documents were fully parsed and processed into an XML representation; each sentence was evaluated against the topic using string, syntax, and semantic characteristics.

clr04n2¶

Run ID: clr04n2
Participant: clresearch
Track: Novelty
Year: 2004
Submission: 9/13/2004
Task: task2
Run description: Topics and documents were fully parsed and processed into an XML representation; each sentence was evaluated against the topic using string, syntax, and semantic characteristics.

clr04n3h1f1¶

Run ID: clr04n3h1f1
Participant: clresearch
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task3
Run description: Topics and documents were fully parsed and processed into an XML representation; each sentence was evaluated against the topic using string, syntax, and semantic characteristics.

clr04n3h1f2¶

Run ID: clr04n3h1f2
Participant: clresearch
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: Topics and documents were fully parsed and processed into an XML representation; each sentence was evaluated against the topic using string, syntax, and semantic characteristics.

clr04n3h2f1¶

Run ID: clr04n3h2f1
Participant: clresearch
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: Topics and documents were fully parsed and processed into an XML representation; each sentence was evaluated against the topic using string, syntax, and semantic characteristics.

clr04n3h2f2¶

Run ID: clr04n3h2f2
Participant: clresearch
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: Topics and documents were fully parsed and processed into an XML representation; each sentence was evaluated against the topic using string, syntax, and semantic characteristics.

clr04n4¶

Run ID: clr04n4
Participant: clresearch
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task4
Run description: Topics and documents were fully parsed and processed into an XML representation; each sentence was evaluated against the topic using string, syntax, and semantic characteristics.

HIL10¶

Run ID: HIL10
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 8/30/2004
Task: task1
Run description: vector space model.

ICT2OKALCEAP¶

Run ID: ICT2OKALCEAP
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: OKAPI additional with Local Context Expansion

ICT2OKAPIAP¶

Run ID: ICT2OKAPIAP
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/13/2004
Task: task2
Run description: model sentence with OKAPI and detect novel sentences using average previous sentence comparison.

ICT2VSMIG95¶

Run ID: ICT2VSMIG95
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/12/2004
Task: task2
Run description: VSM and novelty detection with information gain.

ICT2VSMLCE¶

Run ID: ICT2VSMLCE
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/13/2004
Task: task2
Run description: We empoly dynamic threshold according to the topic date. And make expansion with local cooccurrence.

ICT2VSMOLP¶

Run ID: ICT2VSMOLP
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/12/2004
Task: task2
Run description: Using Word Overlapping

ICT3OKAPFDBK¶

Run ID: ICT3OKAPFDBK
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: OKAPI model and information gain with relevance feedback

ICT3OKAPIIG¶

Run ID: ICT3OKAPIIG
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: OKAPI model and Information gain-based novelty detection

ICT3OKAPIOLP¶

Run ID: ICT3OKAPIOLP
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: OKAPI model and word overlapping-based novelty detection

ICT3VSMOLP¶

Run ID: ICT3VSMOLP
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: Baseline with SVM and word overlapping

ICT4IG¶

Run ID: ICT4IG
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/22/2004
Task: task4
Run description: relevance degree differetial

ICT4OKAAP¶

Run ID: ICT4OKAAP
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/22/2004
Task: task4
Run description: Detect new information comparing with average previous relevance

ICT4OKAPIIG¶

Run ID: ICT4OKAPIIG
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/22/2004
Task: task4
Run description: relevance degree differetial based on OKAPI model

ICT4OVERLAP¶

Run ID: ICT4OVERLAP
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/21/2004
Task: task4
Run description: Simple word overlapping

ICT4OVLPCHI¶

Run ID: ICT4OVLPCHI
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/22/2004
Task: task4
Run description: Word Overlapping considering word weight

ICTOKAPIOVLP¶

Run ID: ICTOKAPIOVLP
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: Adapt OKAPI IR model on sentence retrieval while use word overlap to predict new sentences.

ICTVSMCOSAP¶

Run ID: ICTVSMCOSAP
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: Relevance retrieval using VSM with Cosine normalization, and novelty detection based on average similarity with previous sentences.

ICTVSMFDBKH¶

Run ID: ICTVSMFDBKH
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: Based on basic Vector Space Model,We introduce pseu-feedback to improve relevance performace.

ICTVSMFDBKL¶

Run ID: ICTVSMFDBKL
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: Based on basic Vector Space Model,We introduce top 20% result as pseu-feedback to improve relevance performace.

ICTVSMLCE¶

Run ID: ICTVSMLCE
Participant: cas.ict.wang
Track: Novelty
Year: 2004
Submission: 9/2/2004
Task: task1
Run description: Basic VSM and adding local highly co-occurrence term expansion.

Irit1T3¶

Run ID: Irit1T3
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: Compared to the other runs for this task, information on relevance is not only used to expand the qurery, but also to add all the relevant sentences from these documents to the retrievd set. The filtering process is the same than in task 1.

Irit2T2¶

Run ID: Irit2T2
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: all new

Irit2Task3¶

Run ID: Irit2Task3
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: term expansion based on relevance feedback

Irit3Task3¶

Run ID: Irit3Task3
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: same as Irit2Task3 but with different parameter values

Irit4Task3¶

Run ID: Irit4Task3
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: same as Irit2Task3 but with different parameter values

Irit5Task3¶

Run ID: Irit5Task3
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: same as Irit2Task3 but with different parameter values

IRITT1¶

Run ID: IRITT1
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 8/30/2004
Task: task1
Run description: Representative terms are filtered according to their tf.idf weights and re-weighted as very relevant, relevant or non relevant.

IRITT2¶

Run ID: IRITT2
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 8/30/2004
Task: task1
Run description: Same type of process than run IRITT1 but with different parameter values.

IRITT3¶

Run ID: IRITT3
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 8/30/2004
Task: task1
Run description: In addition to the process used in IRITT1 run, this run uses blind relevant feedback.

IRITT4¶

Run ID: IRITT4
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: Same approach than IRITT1, with different parameter values

IRITT5¶

Run ID: IRITT5
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: Same approach than IRITT1, with different parameter values

IritTask2¶

Run ID: IritTask2
Participant: irit.sig.boughanem
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: New sentences are filtered based on their similarity with previous proceeded sentences and on their similarity with a virtual sentence built from the best new sentences.

ISIALL04¶

Run ID: ISIALL04
Participant: usc.isi.kim
Track: Novelty
Year: 2004
Submission: 8/12/2004
Task: task1
Run description: All sentences are relevant and new.

ISIRUN204¶

Run ID: ISIRUN204
Participant: usc.isi.kim
Track: Novelty
Year: 2004
Submission: 8/12/2004
Task: task1
Run description: For the opinion topics, we used opinion-bearing words as indicators of relevant sentences and for the event topics, we treated them as document IR.

ISIRUN304¶

Run ID: ISIRUN304
Participant: usc.isi.kim
Track: Novelty
Year: 2004
Submission: 8/12/2004
Task: task1
Run description: For the opinion topics, we used opinion-bearing words as indicators of relevant sentences and for the event topics, we treated them as document IR.

ISIRUN404¶

Run ID: ISIRUN404
Participant: usc.isi.kim
Track: Novelty
Year: 2004
Submission: 8/12/2004
Task: task1
Run description: For the opinion topics, we used opinion-bearing words as indicators of relevant sentences and for the event topics, we treated them as document IR.

ISIRUN504¶

Run ID: ISIRUN504
Participant: usc.isi.kim
Track: Novelty
Year: 2004
Submission: 8/12/2004
Task: task1
Run description: For the opinion topics, we used opinion-bearing words as indicators of relevant sentences and for the event topics, we treated them as document IR.

LRIaze1¶

Run ID: LRIaze1
Participant: u.paris.lri
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: statistical approach based on information (word, term, person) found in topic

LRIaze12¶

Run ID: LRIaze12
Participant: u.paris.lri
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: statistical approach based on a heavy parsing and use of static and dynamic threshold for novelty detection.

LRIaze2¶

Run ID: LRIaze2
Participant: u.paris.lri
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: statistical approach based on information (word, term, person) found in topic and person found in text by coreference resolution

LRIaze22¶

Run ID: LRIaze22
Participant: u.paris.lri
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: statistical approach based on a heavy parsing, coreference resolution and use of static and dynamic threshold for novelty detection.

LRIaze3¶

Run ID: LRIaze3
Participant: u.paris.lri
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: Mixed weighted approach statistical and knowledge based approach

LRIaze32¶

Run ID: LRIaze32
Participant: u.paris.lri
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: mixed approach knowledge and statistical approach based on a heavy parsing and use of static and dynamic threshold for novelty detection.

LRIaze4¶

Run ID: LRIaze4
Participant: u.paris.lri
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: Mixed weighted approach statistical and knowledge based approach

LRIaze42¶

Run ID: LRIaze42
Participant: u.paris.lri
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: mixed approach knowledge and statistical approach based on a heavy parsing and use of static and dynamic threshold for novelty detection.

LRIaze5¶

Run ID: LRIaze5
Participant: u.paris.lri
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: Mixed weighted approach statistical and knowledge based approach

LRIaze52¶

Run ID: LRIaze52
Participant: u.paris.lri
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: mixed approach knowledge and statistical approach based on a heavy parsing and use of static and dynamic threshold for novelty detection.

MeijiHIL1cfs¶

Run ID: MeijiHIL1cfs
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: Relevance, Conceptual fuzzy sets is used for term expanstion. Novelty, We used Sentence weight score, redundancy score and scarcity score.

MeijiHIL1odp¶

Run ID: MeijiHIL1odp
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: Relevance, ODP is used for term expanstion. Novelty, We used Sentence weight score, redundancy score and scarcity score.

MeijiHIL2CS¶

Run ID: MeijiHIL2CS
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: We used redundancy score using term expansion by conceptual fuzzy sets and in considaration of the dupulication and the freshness of a sentence.

MeijiHIL2RS¶

Run ID: MeijiHIL2RS
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/13/2004
Task: task2
Run description: We used redundancy score and scarcity score in considaration of the dupulication and the freshness of a sentence.

MeijiHIL2WCS¶

Run ID: MeijiHIL2WCS
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: We used sentence weight score and redundancy score using term expansion by conceptual fuzzy sets and in considaration of the N-window-tf-idf and the dupulication and the freshness of a sentence.

MeijiHIL2WR¶

Run ID: MeijiHIL2WR
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/13/2004
Task: task2
Run description: We used sentence weight score and redundancy score and in considaration of the N-window-tf-idf and the dupulication of a sentence.

MeijiHIL2WRS¶

Run ID: MeijiHIL2WRS
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/13/2004
Task: task2
Run description: We used sentence weight score and redundancy score and scarcity score in considaration of the N-window-tf-idf and the dupulication and the freshness of a sentence.

MeijiHIL3¶

Run ID: MeijiHIL3
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/13/2004
Task: task3
Run description: Vector space model is used. In Novelty, we used sentence weight score, redundancy score and scarcity score.

MeijiHIL3Tc¶

Run ID: MeijiHIL3Tc
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/13/2004
Task: task3
Run description: Vector space model. In Relevance, Conceptual Fuzzy Sets is used for Topic and Sentences expansion. In Novelty, we used sentence weight score, redundancy score and scarcity score.

MeijiHIL3TSc¶

Run ID: MeijiHIL3TSc
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/13/2004
Task: task3
Run description: Vector space model. In Relevance, Conceptual Fuzzy Sets is used for Topic expansion. In Novelty, we used sentence weight score, redundancy score and scarcity score.

MeijiHIL4RS¶

Run ID: MeijiHIL4RS
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/22/2004
Task: task4
Run description: We used redundancy score and scarcity score in consideration of the dupulication and the freshness of a sentence.

MeijiHIL4RSc¶

Run ID: MeijiHIL4RSc
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/22/2004
Task: task4
Run description: We used redundancy score using term expansion by conceptual fuzzy sets and scarcity score in consideration of the dupulication and the freshness of a sentence.

MeijiHIL4WR¶

Run ID: MeijiHIL4WR
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/21/2004
Task: task4
Run description: We used sentence weight score and redundancy score and in considerasion of the N-wondow-idf and the dupulication of a sentence.

MeijiHIL4WRc¶

Run ID: MeijiHIL4WRc
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/21/2004
Task: task4
Run description: We used sentence weight score and redundancy score using term expansion by conceptual fuzzy sets in considerasion of the N-window-idf and the dupulication.

MeijiHIL4WRS¶

Run ID: MeijiHIL4WRS
Participant: meiji.u
Track: Novelty
Year: 2004
Submission: 9/21/2004
Task: task4
Run description: We used sentence weight score and redundancy score and scarcity score in considerasion of the N-wondow-idf and the dupulication and the freshness of a sentence.

novcolp1¶

Run ID: novcolp1
Participant: columbia.u.schiffman
Track: Novelty
Year: 2004
Submission: 9/9/2004
Task: task2
Run description: Basically we tried to segment the documents into new and old sections and accepted runs of sentences.

novcolp2¶

Run ID: novcolp2
Participant: columbia.u.schiffman
Track: Novelty
Year: 2004
Submission: 9/9/2004
Task: task2
Run description: We tried to segment the articles into new and old sections and accepted runs of sentences (1 or more).

novcolrcl¶

Run ID: novcolrcl
Participant: columbia.u.schiffman
Track: Novelty
Year: 2004
Submission: 9/9/2004
Task: task2
Run description: We tried to segment the articles into new and old sections and accepted runs of sentences, 1 or more in length.

novcombo¶

Run ID: novcombo
Participant: columbia.u.schiffman
Track: Novelty
Year: 2004
Submission: 9/9/2004
Task: task2
Run description: This run is a combination of one of the segmentation runs and the vector-space run.

novcosine¶

Run ID: novcosine
Participant: columbia.u.schiffman
Track: Novelty
Year: 2004
Submission: 9/9/2004
Task: task2
Run description: Straightforward vector space model with cosine distance to measure similarity.

NTU11¶

Run ID: NTU11
Participant: ntu.chen
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: use reference corpus to identify sentences

NTU12¶

Run ID: NTU12
Participant: ntu.chen
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: use reference corpus to identify sentences

NTU13¶

Run ID: NTU13
Participant: ntu.chen
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: use reference corpus to identify sentences

NTU14¶

Run ID: NTU14
Participant: ntu.chen
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: use reference corpus to identify sentences

NTU15¶

Run ID: NTU15
Participant: ntu.chen
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: use reference corpus to identify sentences

NTU21¶

Run ID: NTU21
Participant: ntu.chen
Track: Novelty
Year: 2004
Submission: 9/9/2004
Task: task2
Run description: use reference corpus to identify novel sentences

NTU22¶

Run ID: NTU22
Participant: ntu.chen
Track: Novelty
Year: 2004
Submission: 9/9/2004
Task: task2
Run description: use reference corpus to identify novel sentences

NTU23¶

Run ID: NTU23
Participant: ntu.chen
Track: Novelty
Year: 2004
Submission: 9/9/2004
Task: task2
Run description: use reference corpus to identify novel sentences

NTU24¶

Run ID: NTU24
Participant: ntu.chen
Track: Novelty
Year: 2004
Submission: 9/9/2004
Task: task2
Run description: use reference corpus to identify novel sentences

NTU25¶

Run ID: NTU25
Participant: ntu.chen
Track: Novelty
Year: 2004
Submission: 9/9/2004
Task: task2
Run description: use reference corpus to identify novel sentences

THUIRnv0411¶

Run ID: THUIRnv0411
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: baseline long query sentence retrieval, selected pool for sentence novelty judgement.

THUIRnv0412¶

Run ID: THUIRnv0412
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: long query sentence retrieval with local QE, named-entity parsing, relevant threshold based on top sentence RSV, selected pool for sentence novelty judgement.

THUIRnv0413¶

Run ID: THUIRnv0413
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: long query sentence retrieval with local QE, named-entity parsing, combine sentence and document RSV, relevant threshold based on top result RSV, selected pool for sentence novelty judgement.

THUIRnv0414¶

Run ID: THUIRnv0414
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: long query sentence retrieval with local QE, named-entity parsing, combine sentence and document RSV, cut result by proportion, selected pool for sentence novelty judgement.

THUIRnv0415¶

Run ID: THUIRnv0415
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: long query sentence retrieval with local QE, filter irrelevant documents according to sentence retrieval result, selected pool for sentence novelty judgement.

THUIRnv0421¶

Run ID: THUIRnv0421
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: overlap method with a tightness restriction on the overlaping words.

THUIRnv0422¶

Run ID: THUIRnv0422
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: previously appeared sentences are selected to a pool, according to their similarity to the current sentence to be judged; a tightness restriction is used in the selection procedure; an overlap comparison between current sentence and the pool is employed.

THUIRnv0423¶

Run ID: THUIRnv0423
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: PCA dimension reduction and sentence vector cosine similarity comparison.

THUIRnv0424¶

Run ID: THUIRnv0424
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: previously appeared sentences are selected to a pool, according to their similarity to the current sentence to be judged; an overlap comparison between current sentence and the pool is employed.

THUIRnv0425¶

Run ID: THUIRnv0425
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: Only nouns, verbs and adjectives are used as input to the retrieval system, same pool overlap tightness method is used to rid of redundant sentences.

THUIRnv0431¶

Run ID: THUIRnv0431
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: long query sentence retrieval, named-entity parsing, combine sentence and document RSV, relevant threshold based on top result RSV, selected pool with tightness restriction for sentence novelty judgement.

THUIRnv0432¶

Run ID: THUIRnv0432
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: long query sentence retrieval with local feedback according to judgement top5 relevants; PCA dimension reduction for sentence vectors and relevant threshold based on top result's cosine similarity to the query are used for retrieval, selected pool with tightness restriction for sentence novelty judgement.

THUIRnv0433¶

Run ID: THUIRnv0433
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: long query sentence retrieval with local feedback according to judgement top5 relevants and named entity parsing; PCA dimension reduction for sentence vectors and relevant threshold based on top result's cosine similarity to the query are used for retrieval, selected pool with tightness restriction for sentence novelty judgement.

THUIRnv0434¶

Run ID: THUIRnv0434
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: Relevant step SVM trained using top5 relevant and first retrieval results of long query. Novelty step the same selected pool with tightness restriction.

THUIRnv0435¶

Run ID: THUIRnv0435
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: long query retrieval result filtering based on sentence similarity. novelty step used selected pool with tightness restriction.

THUIRnv0441¶

Run ID: THUIRnv0441
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/19/2004
Task: task4
Run description: selected pool method with tightness restriction.

THUIRnv0442¶

Run ID: THUIRnv0442
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/19/2004
Task: task4
Run description: sentence-sentence overlap method with tightness restriction.

THUIRnv0443¶

Run ID: THUIRnv0443
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/19/2004
Task: task4
Run description: sentence-sentence overlap method (baseline).

THUIRnv0444¶

Run ID: THUIRnv0444
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/20/2004
Task: task4
Run description: PCA dimension reduction and cosine similarity sentence to sentence comparison.

THUIRnv0445¶

Run ID: THUIRnv0445
Participant: tsinghua.ma
Track: Novelty
Year: 2004
Submission: 9/20/2004
Task: task4
Run description: PCA dimension reduction (in top 5 documents only novel sentences are included) and cosine similarity sentence to sentence comparison.

UIowa04Nov11¶

Run ID: UIowa04Nov11
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: baseline run from 2003 with minor improvements. Novelty is judged by appearance of new named entities or noun phrases. Noun phrases are handled as simple text.

UIowa04Nov12¶

Run ID: UIowa04Nov12
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: Novelty is judged by appearance of new named entities or noun phrases. Noun phrases are expanded into all matching synset IDs using WordNet. Multi-word nounphrases are also expanded to separate nouns and also expanded to synset IDs.

UIowa04Nov13¶

Run ID: UIowa04Nov13
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: Novelty is judged by appearance of new named entities or noun phrases. Noun phrases are sense disambiguated using a local ensemble-based word sense disambiguation compoment. Multi-word nounphrases are also expanded to separate nouns and sense disambiguated.

UIowa04Nov14¶

Run ID: UIowa04Nov14
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 8/30/2004
Task: task1
Run description: Relevant sentences are judged by their similarity to the topic and novel sentences are judged by their similarity to the dynamic knowledge pool.

UIowa04Nov15¶

Run ID: UIowa04Nov15
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 8/30/2004
Task: task1
Run description: Relevant sentences are judged by their similarity to the topic and novel sentences are judged by their similarity to the dynamic knowledge pool.

UIowa04Nov21¶

Run ID: UIowa04Nov21
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: Named entities and noun phrases are used as triggers for novelty. We expand NPs to all matching synset IDs and use the IDs for comparison.

UIowa04Nov22¶

Run ID: UIowa04Nov22
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task2
Run description: Named entities and noun phrases are used as triggers for novelty. We sense-disambiguate NPs (and their constituents) and use the synset ID for the sense for comparison.

UIowa04Nov23¶

Run ID: UIowa04Nov23
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: If the similarity between the sentence and the dynamic knowledge pool is below a certain threshold, the sentence is retrieved as novel.

UIowa04Nov24¶

Run ID: UIowa04Nov24
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: If the similarity between the sentence and the dynamic knowledge pool is below a certain threshold, the sentence is retrieved as novel.

UIowa04Nov25¶

Run ID: UIowa04Nov25
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: If the similarity between the sentence and the dynamic knowledge pool is below a certain threshold, the sentence is retrieved as novel.

UIowa04Nov31¶

Run ID: UIowa04Nov31
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task3
Run description: Named entities and noun phrases are used as triggers for novelty. We expand NPs to all matching synset IDs and use the IDs for comparison.

UIowa04Nov32¶

Run ID: UIowa04Nov32
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task3
Run description: Named entities and noun phrases are used as triggers for novelty. We sense-disambiguate NPs (and their constituents) and use the synset ID for the sense for comparison.

UIowa04Nov33¶

Run ID: UIowa04Nov33
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: Use the provided relevant and novel sentences to train the best relevant and novel thresholds.

UIowa04Nov34¶

Run ID: UIowa04Nov34
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: Use the provided relevant and novel sentences to train the best relevant and novel thresholds.

UIowa04Nov35¶

Run ID: UIowa04Nov35
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: Use the provided relevant and novel sentences to train the best relevant and novel thresholds.

UIowa04Nov41¶

Run ID: UIowa04Nov41
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task4
Run description: Named entities and noun phrases are used as triggers for novelty. We expand NPs to all matching synset IDs and use the IDs for comparison.

UIowa04Nov42¶

Run ID: UIowa04Nov42
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/14/2004
Task: task4
Run description: Named entities and noun phrases are used as triggers for novelty. We sense-disambiguate NPs (and their constituents) and use the synset ID for the sense for comparison.

UIowa04Nov43¶

Run ID: UIowa04Nov43
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/22/2004
Task: task4
Run description: Get the optimal novel retrieval threshold by the provided novel sentences.

UIowa04Nov44¶

Run ID: UIowa04Nov44
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/22/2004
Task: task4
Run description: Get the optimal novel retrieval threshold by the provided novel sentences.

UIowa04Nov45¶

Run ID: UIowa04Nov45
Participant: u.iowa
Track: Novelty
Year: 2004
Submission: 9/22/2004
Task: task4
Run description: Get the optimal novel retrieval threshold by the provided novel sentences.

umich0411¶

Run ID: umich0411
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: MaxEnt training on the 2003 data with features Centroid, Length and QueryTitleCosine to find the relevant sentences; use a similarity threshold to find the new sentences.

umich0412¶

Run ID: umich0412
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 8/31/2004
Task: task1
Run description: MaxEnt training on the 2003 data with features Centroid, Length and QueryTitleCosine to find the relevant sentences; use a similarity threshold to find the new sentences.

umich0413¶

Run ID: umich0413
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: MaxEnt training on the 2003 data with features Centroid and QueryDescWordOverlap to find the relevant sentences; use a similarity threshold to find the new sentences.

umich0414¶

Run ID: umich0414
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: MaxEnt training on the 2003 data with features LexPageRank and QueryDescWordOverlap to find the relevant sentences; use a similarity threshold to find the new sentences.

umich0415¶

Run ID: umich0415
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/1/2004
Task: task1
Run description: MaxEnt training on the 2003 data with features Length and QueryTitleWordOverlap to find the relevant sentences; use a similarity threshold to find the new sentences.

umich0421¶

Run ID: umich0421
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: We formed a graph of sentences connected according to a cosine similarity threshold (=0.7), then applied the PageRank algorithm

umich0422¶

Run ID: umich0422
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: We formed a graph of sentences connected according to a cosine similarity threshold (=0.7), then applied the PageRank algorithm

umich0423¶

Run ID: umich0423
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: We formed a graph of sentences connected according to a cosine similarity threshold (=0.9), then applied the PageRank algorithm

umich0424¶

Run ID: umich0424
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: We formed a graph of sentences connected according to a cosine similarity threshold (=0.5), then applied the PageRank algorithm

umich0425¶

Run ID: umich0425
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task2
Run description: We formed a graph of sentences connected according to a cosine similarity threshold (=0.8), then applied the PageRank algorithm

umich0431¶

Run ID: umich0431
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: Given the relevant sentences, we expanded the "description" and tried to eliminate the irrelevant documents by looking at word overlap. Finding new sentences is the same as in Task 2.

umich0432¶

Run ID: umich0432
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: Given the relevant sentences, we expanded the "description" and tried to eliminate the irrelevant documents by looking at word overlap. Finding new sentences is the same as in Task 2.

umich0433¶

Run ID: umich0433
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/15/2004
Task: task3
Run description: Given the relevant sentences, we expanded the "description" and tried to eliminate the irrelevant documents by looking at word overlap. Finding new sentences is the same as in Task 2.

umich0434¶

Run ID: umich0434
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: Given the relevant sentences, we expanded the "description" and tried to eliminate the irrelevant documents by looking at word overlap. Finding new sentences is the same as in Task 2.

umich0435¶

Run ID: umich0435
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/16/2004
Task: task3
Run description: Given the relevant sentences, we expanded the "description" and tried to eliminate the irrelevant documents by looking at word overlap. Finding new sentences is the same as in Task 2.

umich0441¶

Run ID: umich0441
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/21/2004
Task: task4
Run description: Same as in task 2. Howeever, we used the 2004 data provided by NIST for training instead of 2003 data.

umich0442¶

Run ID: umich0442
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/21/2004
Task: task4
Run description: Same as in task 2. Howeever, we used the 2004 data provided by NIST for training instead of 2003 data. (cosine threshold=0.6)

umich0443¶

Run ID: umich0443
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/21/2004
Task: task4
Run description: Same as in task 2. Howeever, we used the 2004 data provided by NIST for training instead of 2003 data. (cosine threshold=0.8)

umich0444¶

Run ID: umich0444
Participant: u.michigan
Track: Novelty
Year: 2004
Submission: 9/21/2004
Task: task4
Run description: Same as in task 2. Howeever, we used the 2004 data provided by NIST for training instead of 2003 data. (cosine threshold=0.4)