Runs - Contextual Suggestion 2015¶

11¶

Run ID: 11
Participant: USI
Track: Contextual Suggestion
Year: 2015
Submission: 8/24/2015
Task: batch
Run description: Users are modeled based on their history using other people's reviews and the place information. For every user a classifier is then trained based on this information. There are also some simple measures for information such as places categories. By combining all these scores, the highest scored places are suggested.

22¶

Run ID: 22
Participant: USI
Track: Contextual Suggestion
Year: 2015
Submission: 8/24/2015
Task: batch
Run description: Users are modeled based on their history using other people's reviews and the place information. For every user a classifier is then trained based on this information. There are also some simple measures for information such as places categories. By combining all these scores, the highest scored places are suggested.

BJUTA¶

Run ID: BJUTA
Participant: BJUT
Track: Contextual Suggestion
Year: 2015
Submission: 8/23/2015
Task: batch
Run description: WeiTong chen,CNMF

BJUTb¶

Run ID: BJUTb
Participant: BJUT
Track: Contextual Suggestion
Year: 2015
Submission: 8/23/2015
Task: batch
Run description: use website content to train a classifier for each user

fr¶

Run ID: fr
Participant: udel_fang
Track: Contextual Suggestion
Year: 2015
Submission: 8/11/2015
Task: batch
Run description: Reviews from Yelp, Tripadvisor, Opentable were collected. The reviews serve as both the basis of user profile and candidate profile. The similarity between user profile and candidate profile is computed as the ranking score. An advantaged context filtering approach is applied to the ranking score and the final score is generated.

IITBHU_1¶

Run ID: IITBHU_1
Participant: DPLAB_IITBHU
Track: Contextual Suggestion
Year: 2015
Submission: 8/19/2015
Task: batch
Run description: Suggestions were given by matching corresponding synsets of rated attractions and candidate suggestions.

IITBHU_2¶

Run ID: IITBHU_2
Participant: DPLAB_IITBHU
Track: Contextual Suggestion
Year: 2015
Submission: 8/19/2015
Task: batch
Run description: clusters based on ratings were formed and candidate suggestions were matched with their corresponding synsets

IRKM1¶

Run ID: IRKM1
Participant: ucsc
Track: Contextual Suggestion
Year: 2015
Submission: 6/20/15
Task: live

IRKM2¶

Run ID: IRKM2
Participant: ucsc
Track: Contextual Suggestion
Year: 2015
Submission: 6/20/15
Task: live

LavalIVA-run1¶

Results | Participants | Input | Summary | Appendix

Run ID: LavalIVA-run1
Participant: LavalIVA
Track: Contextual Suggestion
Year: 2015
Submission: 5/27/2015
Task: live

LavalIVA_1¶

Results | Participants | Input | Summary | Appendix

Run ID: LavalIVA_1
Participant: LavalIVA
Track: Contextual Suggestion
Year: 2015
Submission: 8/23/2015
Task: batch
Run description: This run is powered by an elastic search engine aided with an SVM classifier. Most data are from Yelp downloaded by Scrapy.
Code: https://bitbucket.org/mojians/contextual-suggestion/branch/batch_server

LavalIVA_2¶

Results | Participants | Input | Summary | Appendix

Run ID: LavalIVA_2
Participant: LavalIVA
Track: Contextual Suggestion
Year: 2015
Submission: 8/23/2015
Task: batch
Run description: Run2 is similar with Run1 but take category diversity into consideration.
Code: https://bitbucket.org/mojians/contextual-suggestion/branch/batch_server

nr¶

Run ID: nr
Participant: udel_fang
Track: Contextual Suggestion
Year: 2015
Submission: 8/11/2015
Task: batch
Run description: Reviews from Yelp, Tripadvisor, Opentable were collected. The reviews serve as both the basis of user profile and candidate profile. The similarity between user profile and candidate profile is computed as the ranking score. An advantaged context filtering approach is applied to the ranking score and the final score is generated.

PLM1¶

Run ID: PLM1
Participant: UAmsterdam
Track: Contextual Suggestion
Year: 2015
Submission: 8/7/2015
Task: batch
Run description: We have crawled the TREC 2015 Contextual Suggestion Track collection, and used them to give the suggestions. In this run, we propose a parsimonious personalization approach, which builds a positive parsimonious language model for each profile and rank the suggestions based on the generated profiles.

PLM2¶

Run ID: PLM2
Participant: UAmsterdam
Track: Contextual Suggestion
Year: 2015
Submission: 8/7/2015
Task: batch
Run description: We have crawled the TREC 2015 Contextual Suggestion Track collection, and used them to give the suggestions. In this run, we propose a parsimonious personalization approach, in which we build a positive parsimonious language model and a negative parsimonious language model for each profile. Afterward, we build a new parsimonious language model as the user profile, in which terms having high probability in the negative parsimonious language model are penalized. Based on the generated profile, the suggestions are ranked.

RUN1¶

Run ID: RUN1
Participant: ucsc
Track: Contextual Suggestion
Year: 2015
Submission: 8/22/2015
Task: batch
Run description: We extract the data features, including text features, user features, and domain preference of users, and then train the END model from weka on the training data set. END is a meta classifier for handling multi-class datasets with 2-class classifiers by building an ensemble of nested dichotomies.

RUN2¶

Run ID: RUN2
Participant: ucsc
Track: Contextual Suggestion
Year: 2015
Submission: 8/22/2015
Task: batch
Run description: We extract the data features, including text features, user features, and domain preference of users, and then train the SOM model from Weka on the training data set. SMO implements John Platt's sequential minimal optimization algorithm for training a support vector classifier.

SCIAI_runA¶

Run ID: SCIAI_runA
Participant: Siena_SUCCESS
Track: Contextual Suggestion
Year: 2015
Submission: 7/30/2015
Task: batch
Run description: We separated the candidates from the rated examples into two separate files, and retrieved only the information from the given collection that we needed. We then sent those attractions through our API system to categorize each one. The categories associated with the rated attractions were scored based on attraction rating. Using these category scores, we gave a numeric value to each candidate attraction. For this run, candidates were ranked after passing through our unique penalty function.

SCIAI_runB¶

Run ID: SCIAI_runB
Participant: Siena_SUCCESS
Track: Contextual Suggestion
Year: 2015
Submission: 7/30/2015
Task: batch
Run description: We separated the candidates from the rated examples into two separate files, and retrieved only the information from the given collection that we needed. We then sent those attractions through our API system to categorize each one. The categories associated with the rated attractions were scored based on attraction rating. Using these category scores, we gave a numeric value to each candidate attraction. For this run, candidates were ranked based on numeric value without passing through our penalty function.

TJU_BASE¶

Results | Participants | Input | Summary | Appendix

Run ID: TJU_BASE
Participant: TJU_CSIR
Track: Contextual Suggestion
Year: 2015
Submission: 7/4/2015
Task: live

TJU_CSIR_TOPIC¶

Results | Participants | Input | Summary | Appendix

Run ID: TJU_CSIR_TOPIC
Participant: TJU_CSIR
Track: Contextual Suggestion
Year: 2015
Submission: 8/24/2015
Task: batch
Run description: Our basic idea is to discover the abstract "semantics" that occur in the collection of documents. Another is to use the "tags" to lable the preference document, train the topic model for user and rank the candidate documents.

TJU_CSIR_VSM¶

Results | Participants | Input | Summary | Appendix

Run ID: TJU_CSIR_VSM
Participant: TJU_CSIR
Track: Contextual Suggestion
Year: 2015
Submission: 8/24/2015
Task: batch
Run description: Our basic idea is to discover the abstract "semantics" that occur in the collection of documents. One is to represent all document in the continues vector space learnt by neural network language models (NNML)(in our implementation, we adopt the word2vec model in [Tomas Mikolov, NIPS, 2013]). Then we learn the user profile model to rank candidate documents using cos similarity with preference document.

UDInfoCS2015¶

Run ID: UDInfoCS2015
Participant: udel_fang
Track: Contextual Suggestion
Year: 2015
Submission: 6/22/2015
Task: live

uogTrCSFM¶

Run ID: uogTrCSFM
Participant: uogTr
Track: Contextual Suggestion
Year: 2015
Submission: 8/24/2015
Task: batch
Run description: Factorisation machines with venue features (statistics and categories) from Foursquare.

uogTrCsLtrUDepCat¶

Run ID: uogTrCsLtrUDepCat
Participant: uogTr
Track: Contextual Suggestion
Year: 2015
Submission: 7/8/2015
Task: live

uogTrCsLtrUInd¶

Run ID: uogTrCsLtrUInd
Participant: uogTr
Track: Contextual Suggestion
Year: 2015
Submission: 6/30/2015
Task: live

uogTrCSLVPC¶

Run ID: uogTrCSLVPC
Participant: uogTr
Track: Contextual Suggestion
Year: 2015
Submission: 8/24/2015
Task: batch
Run description: User-independent and user-dependent features extracted from Foursquare, combined using advanced learning to rank. Context-dependent features are also present.

USST1¶

Results | Participants | Input | Summary | Appendix

Run ID: USST1
Participant: USST
Track: Contextual Suggestion
Year: 2015
Submission: 8/5/2015
Task: batch
Run description: Ranking the related website by users'preference

WaterlooRunA¶

Run ID: WaterlooRunA
Participant: WaterlooClarke
Track: Contextual Suggestion
Year: 2015
Submission: 6/25/2015
Task: live

WaterlooRunB¶

Run ID: WaterlooRunB
Participant: WaterlooClarke
Track: Contextual Suggestion
Year: 2015
Submission: 6/29/2015
Task: live