Runs - Contextual Suggestion 2016¶

1¶

Run ID: 1
Participant: FUM-IRLAB
Track: Contextual Suggestion
Year: 2016
Submission: 7/26/2016
Task: phase1
Run description: We used collaborative filtering methods for finding similar users. We used ratings for finingd most popular places in related categories and categories that had been given good rates by similar users.

1st_subminssion¶

Results | Participants | Input | Summary | Appendix

Run ID: 1st_subminssion
Participant: CityUHKGeng
Track: Contextual Suggestion
Year: 2016
Submission: 9/8/2016
Task: phase2
Run description: We use tag-based features to represent the user profile and each candidate. Similarity scores are calculated between user profile and candidate representation. Ranking result is based on the similarity scores.

2¶

Run ID: 2
Participant: FUM-IRLAB
Track: Contextual Suggestion
Year: 2016
Submission: 7/26/2016
Task: phase1
Run description: We used Deep Learning methods for finding similar content related to profiles

3¶

Run ID: 3
Participant: FUM-IRLAB
Track: Contextual Suggestion
Year: 2016
Submission: 7/26/2016
Task: phase1
Run description: We used Deep Learning methods for finding similar content related to profiles

ADAPT_TCD_br1¶

Run ID: ADAPT_TCD_br1
Participant: ADAPT_TCD
Track: Contextual Suggestion
Year: 2016
Submission: 9/8/2016
Task: phase2
Run description: In this approach we rank candidates based on its semantic (ontological) similarity with the users positive model. Our approach is an ontology based approach. We utilize Foursquare categories hierarchy and used it as the ontology for our approach. The approach is based on three models: User Model, Document Model, and a Rule Based Model. In User Model we build positive model based on attractions that the user has rated high and build negative model based on attractions that the user has rated low. Document Model enriches documents with extra metadata from Foursquare and attach concepts to each document from the ontology. The Rule based model is used to give low score for candidates that dont match the rules in that model. These rules are made based on the ontology as well and are modeled based on the users context. Example on the Rule based is that if a users trip duration is Night out, an attraction that belongs to Museum concept (from the ontology) is given a lower rank as Museums are not suitable for a night out. In this run (ADAPT_TCD_br1) we rank the candidates based on the semantic similarity plus we measure the similarity (the overlap) of a candidates tags and positive profiles tags.

ADAPT_TCD_br2¶

Run ID: ADAPT_TCD_br2
Participant: ADAPT_TCD
Track: Contextual Suggestion
Year: 2016
Submission: 9/8/2016
Task: phase2
Run description: In this approach we rank candidates based on its semantic (ontological) similarity with the users positive model. Our approach is an ontology based approach. We utilize Foursquare categories hierarchy and used it as the ontology for our approach. The approach is based on three models: User Model, Document Model, and a Rule Based Model. In User Model we build positive model based on attractions that the user has rated high and build negative model based on attractions that the user has rated low. Document Model enriches documents with extra metadata from Foursquare and attach concepts to each document from the ontology. The Rule based model is used to give low score for candidates that dont match the rules in that model. These rules are made based on the ontology as well and are modeled based on the users context. Example on the Rule based is that if a users trip duration is Night out, an attraction that belongs to Museum concept (from the ontology) is given a lower rank as Museums are not suitable for a night out. In this run (ADAPT_TCD_br2) we rank the candidates based on the semantic similarity plus we measure the similarity (the overlap) of a candidates tags and positive profiles tags. But the difference is that we give a lower rank to candidates that have no metadata from Foursquare.

ADAPT_TCD_br3¶

Run ID: ADAPT_TCD_br3
Participant: ADAPT_TCD
Track: Contextual Suggestion
Year: 2016
Submission: 9/8/2016
Task: phase2
Run description: In this approach we rank candidates based on its semantic (ontological) similarity with the users positive model. Our approach is an ontology based approach. We utilize Foursquare categories hierarchy and used it as the ontology for our approach. The approach is based on three models: User Model, Document Model, and a Rule Based Model. In User Model we build positive model based on attractions that the user has rated high and build negative model based on attractions that the user has rated low. Document Model enriches documents with extra metadata from Foursquare and attach concepts to each document from the ontology. The Rule based model is used to give low score for candidates that dont match the rules in that model. These rules are made based on the ontology as well and are modeled based on the users context. Example on the Rule based is that if a users trip duration is Night out, an attraction that belongs to Museum concept (from the ontology) is given a lower rank as Museums are not suitable for a night out. In this run (ADAPT_TCD_br3) we rank the candidates based only on the semantic similarity and we do not measure the similarity of a candidates tags and positive profiles tags.

ADAPT_TCD_r1¶

Run ID: ADAPT_TCD_r1
Participant: ADAPT_TCD
Track: Contextual Suggestion
Year: 2016
Submission: 7/26/2016
Task: phase1
Run description: Our approach is based on ontology. We utilize Foursquare categories hierarchy and used it as the ontology for our approach. Our approach is based on three models: User Model, Document Model, and a Rule Based Model. We build the three models based on that ontology and attractions are suggested to the user based on the ontological similarity between the attraction (Document Model) and the user (User Model) and the rule based model is built to eliminate attractions that could be selected by the approach but dont match the user context. In the user model, we rank concepts based on users preferences and then select attractions based on these high ranked concepts. In this run (ADAPT_TCD_r1), if the approach couldnt find attractions for a specific concept (foursquare category), it retrieves more of the attractions that belong to the highest ranked concepts.

ADAPT_TCD_r2¶

Run ID: ADAPT_TCD_r2
Participant: ADAPT_TCD
Track: Contextual Suggestion
Year: 2016
Submission: 7/26/2016
Task: phase1
Run description: Our approach is based on ontology. We utilize Foursquare categories hierarchy and used it as the ontology for our approach. Our approach is based on three models: User Model, Document Model, and a Rule Based Model. We build the three models based on that ontology and attractions are suggested to the user based on the ontological similarity between the attraction (Document Model) and the user (User Model) and the rule based model is built to eliminate attractions that could be selected by the approach but dont match the user context. In the user model, we rank concepts based on users preferences and then select attractions based on these high ranked concepts In this Run (ADAPT_TCD_r2), if the approach couldnt find attractions for a specific concept (foursquare category), it looks for a concept that has a high ontological similarity with that concept. We rely on an Edge-counting approach proposed by Wu and Palmer to measure the ontological similarity between concepts.

bupt_runA¶

Run ID: bupt_runA
Participant: bupt_pris_2016
Track: Contextual Suggestion
Year: 2016
Submission: 7/25/2016
Task: phase1
Run description: First, we crawl the information about the attractions on Yelp and Foursquare; Second, we analysis the categories that the user like through preferences; Third, we give the suggestions according the users' gender, age, preferences and location.

CasualChocolate¶

Results | Participants | Input | Summary | Appendix

Run ID: CasualChocolate
Participant: SCIAICLTeam
Track: Contextual Suggestion
Year: 2016
Submission: 9/7/2016
Task: phase2
Run description: This run modeled users and attractions to properly identify relevant information on user-attraction pairs. Using this information, this run scores each pair based on relevancy with an emphasis on common user preferences.

cs.2_.4_max¶

Run ID: cs.2_.4_max
Participant: bupt_pris_2016
Track: Contextual Suggestion
Year: 2016
Submission: 9/6/2016
Task: phase2
Run description: First, we crawled information about all the attractions on yelp, foursquere web site; Second, for each user, we computed the average score of each tag in the tags_list according to the information of the his preference list; Third, we computed tag_sore for each candidate according to the candidate's tags (if the attractions tag list is empty, we will map the category tags crawled to tags in the given tag_list).

cs.3_.3_avg¶

Run ID: cs.3_.3_avg
Participant: bupt_pris_2016
Track: Contextual Suggestion
Year: 2016
Submission: 9/6/2016
Task: phase2
Run description: We considered the duration,group information at this submissions and expected a better results. The steps are: First, we crawled information about all the attractions on yelp, foursquere web site; Second, for each user, we computed the average score of each tag in the tags_list according to the information of the his preference list; Third, we computed tag_sore for each candidate according to the candidate's tags (if the attractions tag list is empty, we will map the category tags crawled to tags in the given tag_list); Fourth, we taked into account the comprehensive consideration of the tag_score and the crawled ratings candidate attractions to get the final score. Fifth, we processed the result based on rules about group, duration and other information.

cs.4_.2_max¶

Run ID: cs.4_.2_max
Participant: bupt_pris_2016
Track: Contextual Suggestion
Year: 2016
Submission: 9/6/2016
Task: phase2
Run description: In this submission, we took into the ratings we crawled from the Internet. The steps as follows: First, we crawled information about all the attractions on yelp, foursquere web site; Second, for each user, we computed the average score of each tag in the tags_list according to the information of the his preference list; Third, we computed tag_sore for each candidate according to the candidate's tags (if the attractions tag list is empty, we will map the category tags crawled to tags in the given tag_list); Fourth, we taked into account the comprehensive consideration of the tag_score and the crawled ratings candidate attractions to get the final score.

DUTH_bcf¶

Run ID: DUTH_bcf
Participant: DUTH
Track: Contextual Suggestion
Year: 2016
Submission: 9/6/2016
Task: phase2
Run description: The results of the run "DUTH_bcf" were computed using the Borda Count voting system as a way to apply data fusion between our other two submitted runs, namely "DUTH_knn" and "DUTH_rocchio".

DUTH_knn¶

Run ID: DUTH_knn
Participant: DUTH
Track: Contextual Suggestion
Year: 2016
Submission: 9/6/2016
Task: phase2
Run description: Using the logic behind the KNN (k-nearest neighbours) algorithm for classification, we assigned a score to each candidate venue based on a weighted average of the ratings of the k (k=7) semantically nearest venues in the user's preferences. We then sorted the candidates from highest to lowest score.

DUTH_rocchio¶

Run ID: DUTH_rocchio
Participant: DUTH
Track: Contextual Suggestion
Year: 2016
Submission: 9/6/2016
Task: phase2
Run description: Using a variation of Rocchio's formula for relevance feedback, we constructed a query of weighted terms for each request according to the ratings of the user's preferences. We then submitted each query to an index of venues of the same location as the request and received a sorted list of the candidates according to the scores returned.

IAPLab1¶

Results | Participants | Input | Summary | Appendix

Run ID: IAPLab1
Participant: IAPLab
Track: Contextual Suggestion
Year: 2016
Submission: 7/25/2016
Task: phase1
Run description: use the tags as feature and use the SVM classifier to get the result

IAPLab2¶

Results | Participants | Input | Summary | Appendix

Run ID: IAPLab2
Participant: IAPLab
Track: Contextual Suggestion
Year: 2016
Submission: 9/7/2016
Task: phase2
Run description: use the svm classier

iitbhu01¶

Run ID: iitbhu01
Participant: DPLAB_IITBHU
Track: Contextual Suggestion
Year: 2016
Submission: 9/8/2016
Task: phase2
Run description: The complete algorithm is based on similarity of user's taste and the candidate suggestions reflected in the form of tags. The tag similarity was computed with the help of relevant wikipedia pages.

iitbhu04¶

Run ID: iitbhu04
Participant: DPLAB_IITBHU
Track: Contextual Suggestion
Year: 2016
Submission: 9/8/2016
Task: phase2
Run description: This run too takes into account the tag similarity between the candidates and user visited attractions, the only difference being for this run we have considered user provided tags along with the ones generated by us.

iitbhu05¶

Run ID: iitbhu05
Participant: DPLAB_IITBHU
Track: Contextual Suggestion
Year: 2016
Submission: 9/8/2016
Task: phase2
Run description: This run too takes into account the tag similarity between the candidates and user visited attractions, the only difference being for this run we have considered user provided tags along with the ones generated by us. This method deviates from the other two runs in the fact that we have considered simple frequency of each tags instead of its weights.

Laval_batch_1¶

Run ID: Laval_batch_1
Participant: LavalLakehead
Track: Contextual Suggestion
Year: 2016
Submission: 9/8/2016
Task: phase2
Run description: A global regressor of common interests from all users

Laval_batch_2¶

Run ID: Laval_batch_2
Participant: LavalLakehead
Track: Contextual Suggestion
Year: 2016
Submission: 9/9/2016
Task: phase2
Run description: word2vec method
Code: https://bitbucket.org/mojians/contextual-suggestion

Laval_batch_3¶

Run ID: Laval_batch_3
Participant: LavalLakehead
Track: Contextual Suggestion
Year: 2016
Submission: 9/9/2016
Task: phase2
Run description: word2vec + global regressor combined
Code: https://bitbucket.org/mojians/contextual-suggestion

Laval_run1¶

Run ID: Laval_run1
Participant: LavalLakehead
Track: Contextual Suggestion
Year: 2016
Submission: 7/27/2016
Task: phase1
Run description: simple term frequency+contextual word embeddings
Code: https://mojians@bitbucket.org/mojians/contextual-suggestion.git

phase2_1¶

Run ID: phase2_1
Participant: FUM-IRLAB
Track: Contextual Suggestion
Year: 2016
Submission: 9/5/2016
Task: phase2
Run description: Using Word2Vec for finding similar places to user interests

phase2_2¶

Run ID: phase2_2
Participant: FUM-IRLAB
Track: Contextual Suggestion
Year: 2016
Submission: 9/5/2016
Task: phase2
Run description: Using collaborative filtering methods on data obtained from ratings in Yelp, Tripadvisor, Foursquare

response_tags¶

Run ID: response_tags
Participant: ExPoSe
Track: Contextual Suggestion
Year: 2016
Submission: 8/1/2016
Task: phase1
Run description: The idea of this run is based on the general idea of "significant words language models" (Mostafa Dehghani, H. Azarbonyad, J. Kamps, D. Hiemstra, and M. Marx. "Luhn Revisited: Significant Words Language Models", To be appeared in the in the proceedings of The ACM International Conference on Information and Knowledge Management (CIKM'16), 2016.), applying on tags.
Code: https://github.com/MostafaDehghani/ContextualSuggestion2016

run_all¶

Run ID: run_all
Participant: ExPoSe
Track: Contextual Suggestion
Year: 2016
Submission: 8/1/2016
Task: phase1
Run description: The idea of this run is based on the general idea of "significant words language models" (Mostafa Dehghani, H. Azarbonyad, J. Kamps, D. Hiemstra, and M. Marx. "Luhn Revisited: Significant Words Language Models", To be appeared in the in the proceedings of The ACM International Conference on Information and Knowledge Management (CIKM'16), 2016.), applying on all the information in the data including tags, page content and contextual information.
Code: https://github.com/MostafaDehghani/ContextualSuggestion2016

run_content¶

Run ID: run_content
Participant: ExPoSe
Track: Contextual Suggestion
Year: 2016
Submission: 8/1/2016
Task: phase1
Run description: The idea of this run is based on the general idea of "significant words language models" (Mostafa Dehghani, H. Azarbonyad, J. Kamps, D. Hiemstra, and M. Marx. "Luhn Revisited: Significant Words Language Models", To be appeared in the in the proceedings of The ACM International Conference on Information and Knowledge Management (CIKM'16), 2016.), applying on page content.
Code: https://github.com/MostafaDehghani/ContextualSuggestion2016

SassyStrawberry¶

Results | Participants | Input | Summary | Appendix

Run ID: SassyStrawberry
Participant: SCIAICLTeam
Track: Contextual Suggestion
Year: 2016
Submission: 9/7/2016
Task: phase2
Run description: This run modeled users and attractions to properly identify relevant information on user-attraction pairs. Using this information, this run scores each pair based on relevancy with an emphasis on specific user preferences.

SWLM¶

Run ID: SWLM
Participant: ExPoSe
Track: Contextual Suggestion
Year: 2016
Submission: 9/12/2016
Task: phase2
Run description: It is based on the idea of "significant words language models" (Mostafa Dehghani, H. Azarbonyad, J. Kamps, D. Hiemstra, and M. Marx. "Luhn Revisited: Significant Words Language Models", To be appeared in the in the proceedings of The ACM International Conference on Information and Knowledge Management (CIKM'16), 2016.), applying on all the information in the data including tags, page content and contextual information.
Code: https://github.com/MostafaDehghani/ContextualSuggestion2016

UAmsterdam1¶

Run ID: UAmsterdam1
Participant: UAmsterdam
Track: Contextual Suggestion
Year: 2016
Submission: 7/27/2016
Task: phase1
Run description: In this run, we indexed TREC CS Web Corpus. In order to build user profiles, we used Parsimonious User Profiling. Finally, we calculated the relevance of each suggestion to the profile using KL-divergence.

UAmsterdam2¶

Run ID: UAmsterdam2
Participant: UAmsterdam
Track: Contextual Suggestion
Year: 2016
Submission: 8/1/2016
Task: phase1
Run description: In this run, we indexed TREC CS Web Corpus. In order to find relevant suggestions to each context, we learnt a contextual tag prediction model using Neural Net. Then we learnt a model of the profile tag preferences using Neural Net. Using both the contextual tag prediction and user tag preference model, we recommend suggestions related to each request.

UAmsterdamCB¶

Run ID: UAmsterdamCB
Participant: UAmsterdam
Track: Contextual Suggestion
Year: 2016
Submission: 9/9/2016
Task: phase2
Run description: The user profile is represented by tags and terms in documents preferences. We propose to compute a weight for each term in the document preferences using word2vec. This weight is based on the similarity between the related tags (document tags). Then, the weight of each term is: term frequency of the term * similarity between term and tag. At last, we have used the content based filtering approach to recommend suggestions based on the learnt profiles.

UAmsterdamDL¶

Run ID: UAmsterdamDL
Participant: UAmsterdam
Track: Contextual Suggestion
Year: 2016
Submission: 9/9/2016
Task: phase2
Run description: We have used Neural Nets to learn how to predict relevant suggestions to the users using tag representations.

uogTrCs¶

Run ID: uogTrCs
Participant: uogTr
Track: Contextual Suggestion
Year: 2016
Submission: 9/7/2016
Task: phase2
Run description: We exploit word embeddings to extract user and venue features to train LTR model.

uogTrCsContext¶

Run ID: uogTrCsContext
Participant: uogTr
Track: Contextual Suggestion
Year: 2016
Submission: 9/7/2016
Task: phase2
Run description: We exploit word embeddings to extract user, venue and context features to train LTR model.

USI1¶

Run ID: USI1
Participant: USI
Track: Contextual Suggestion
Year: 2016
Submission: 7/27/2016
Task: phase1
Run description: Using a similarity score on Foursquare categories

USI2¶

Run ID: USI2
Participant: USI
Track: Contextual Suggestion
Year: 2016
Submission: 7/27/2016
Task: phase1
Run description: Using a similarity score on Foursquare categories and keywords

USI3¶

Run ID: USI3
Participant: USI
Track: Contextual Suggestion
Year: 2016
Submission: 9/8/2016
Task: phase2
Run description: Different features such as categories and tags are fed into FFM.

USI4¶

Run ID: USI4
Participant: USI
Track: Contextual Suggestion
Year: 2016
Submission: 9/9/2016
Task: phase2
Run description: Combination of scores from different sources and information.

USI5¶

Run ID: USI5
Participant: USI
Track: Contextual Suggestion
Year: 2016
Submission: 9/9/2016
Task: phase2
Run description: Combination of scores from different sources and information.

VerbatimVanilla¶

Results | Participants | Input | Summary | Appendix

Run ID: VerbatimVanilla
Participant: SCIAICLTeam
Track: Contextual Suggestion
Year: 2016
Submission: 9/7/2016
Task: phase2
Run description: This run suggests attractions in the same order of the input, verbatim.