Skip to content

Runs - Contextual Suggestion 2012

baselineA

Participants | Proceedings | Input | Appendix

  • Run ID: baselineA
  • Participant: csiro
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: 8a9f8a6aa19cc3d3932abcdb3e5133c2
  • Run description: "Commercial" baseline -- suggestions from a commercial service, minimal filtering, ordered by that service's ratings.

baselineB

Participants | Proceedings | Input | Appendix

  • Run ID: baselineB
  • Participant: csiro
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: e7e92984d8469018427d9c845b9f62a7
  • Run description: This is the "pub run" baseline -- popular pubs and restaurants, as reported by Google, and sorted by ratings.

csiroht

Participants | Proceedings | Input | Appendix

  • Run ID: csiroht
  • Participant: csiro
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/10/2012
  • Type: both
  • Task: main
  • MD5: 75d57f8848890ee4e90a51ad7ea66fe2
  • Run description: Location-appropriate suggestions from a commercial API, ranked by (a) how popular this type of place is at the appropriate time with a small admixture of (b) textual similarity to preference data

csiroth

Participants | Proceedings | Input | Appendix

  • Run ID: csiroth
  • Participant: csiro
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/10/2012
  • Type: both
  • Task: main
  • MD5: 8c1d805d80ce8dd048a86ba58fed7f20
  • Run description: Location-appropriate suggestions from a commercial API, ranked by (a) textual similarity to preference data with a small admixture of (b) how popular this type of place is at the appropriate time

FASILKOMUI01

Participants | Input | Appendix

  • Run ID: FASILKOMUI01
  • Participant: FASILKOMUI
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: 6c3945c92446a890b9127610b026ccdf
  • Run description: We build the user preferences model according to the examples provided. In order to incorporate the contextual info such as geotemporal attributes we gather the info from several popular websites and web service APIs . We also try to find suitable context for a particular place by extracting user comments and examine whether there are lexicons that indicates day, season, time etc. We produce suggestions based on the preference model, after that we filter them by location, context, and then sort it by its score to produce a ranked list of suggestions. Diversity is applied for this run.

FASILKOMUI02

Participants | Input | Appendix

  • Run ID: FASILKOMUI02
  • Participant: FASILKOMUI
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: c81fda3bc1ce70fe0d07f8b5f75cf702
  • Run description: We build the user preferences model according to the examples provided. In order to incorporate the contextual info such as geotemporal attributes we gather the info from several popular websites and web service APIs . We also try to find suitable context for a particular place by extracting user comments and examine whether there are lexicons that indicates day, season, time etc. We produce suggestions based on the preference model, after that we filter them by location, context, and then sort it by its score to produce a ranked list of suggestions. Diversity is applied for this run.

gufinal

Participants | Proceedings | Input | Appendix

  • Run ID: gufinal
  • Participant: Georgetown
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/10/2012
  • Type: both
  • Task: main
  • MD5: ee9a9036140c93efb590090a70b0cfcd
  • Run description: Results were obtained from several different search engines and merged. User preferences were computed using final judgments.

guinit

Participants | Proceedings | Input | Appendix

  • Run ID: guinit
  • Participant: Georgetown
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/10/2012
  • Type: both
  • Task: main
  • MD5: 44646edd882183c68d0ae8a25f370a35
  • Run description: Results were obtained from several different search engines and merged. User preferences were computed using initial judgments.

hplcranking

Participants | Input | Appendix

  • Run ID: hplcranking
  • Participant: HPLC
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: d32911a4c3957b7ec6748fc43f16fea3
  • Run description: We use pairwise ranking based approach to produce the ranked list.

hplcrating

Participants | Input | Appendix

  • Run ID: hplcrating
  • Participant: HPLC
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: e354df9adbe1564317ea34de165aa0ff
  • Run description: We use collaborative filtering based approach to produce the ranked list.

ICTCONTEXTRUN1

Participants | Proceedings | Input | Appendix

  • Run ID: ICTCONTEXTRUN1
  • Participant: ICTNET
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 7/28/2012
  • Type: both
  • Task: main
  • MD5: c067bbc567247bddcff45ea8c20d53f5
  • Run description: The data based on yelp's api and our crawler for the open web. For each context/profile pair we rank the suggested places only by distance.

ICTCONTEXTRUN2

Participants | Proceedings | Input | Appendix

  • Run ID: ICTCONTEXTRUN2
  • Participant: ICTNET
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/1/2012
  • Type: both
  • Task: main
  • MD5: f358e8f10f4950a5e30728738d0bba43
  • Run description: Based on run1 result, we considered using distance, official website or not, personal preference, diversity as ranking metrics. And re-generate place description with four methods.

iritSplit3CPv1

Participants | Proceedings | Input | Appendix

  • Run ID: iritSplit3CPv1
  • Participant: IRIT
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 7/26/2012
  • Type: both
  • Task: main
  • MD5: 6640d0629b39e92be18e7518c2d7dead
  • Run description: This run's suggestions were developped using Google Places API and computing similarity between results and global positive and negative profiles.

iritSplit3CPv2

Participants | Proceedings | Input | Appendix

  • Run ID: iritSplit3CPv2
  • Participant: IRIT
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 7/27/2012
  • Type: both
  • Task: main
  • MD5: 2fd115ae76b22de3ad6c482d5fcee1d8
  • Run description: This run's suggestions were developped using Google Places API and computing similarity between results and positive and negative examples.

PRISabc

Participants | Proceedings | Input | Appendix

  • Run ID: PRISabc
  • Participant: PRIS
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: 9ddf75afb021d4eaa3e02d7502c513cb
  • Run description: mianly use tf-idf and cosine similarity

run01TI

Participants | Proceedings | Input | Appendix

  • Run ID: run01TI
  • Participant: TNO_RadboudUniv
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: 4c22f582fce2592bcf00c7c199d83916
  • Run description: These results were obtained by gathering suggestions through Google Places. Profiles were based on TFIDF values of the terms in the positive examples. For each suggestion the cosine similarity to this "positive" profile was calculated. Discounts were given to items that were more similar to the "negative" profile. Rankings were obtained in this way, and mixed with rankings based on google rating, original google order and a priori preference for a category (based on the categories of the positive examples)

run02K

Participants | Proceedings | Input | Appendix

  • Run ID: run02K
  • Participant: TNO_RadboudUniv
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: f84a08dd17584e3989717a08b2fb66cb
  • Run description: These results were obtained by gathering suggestions through Google Places. Profiles were based on Kullback Leibner Divergence scores of the terms in the positive examples. For each suggestion a ranking was based on the sum of the KL-div scores of the words in the suggestions (only words that occur in the examples are taken into account). These ranks were mixed with rankings based on google rating, original google order and a priori preference for a category (based on the categories of the positive examples)

UAmsCS12wtSUM

Participants | Proceedings | Input | Appendix

  • Run ID: UAmsCS12wtSUM
  • Participant: UAmsterdam
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/8/2012
  • Type: both
  • Task: main
  • MD5: 4a3dfbdb8c239a075bd90ac5dacd7a57
  • Run description: Suggestions are derived from Wikitravel, ranked using a standard language model with descriptions of positive examples as queries. Score for suggestions for individual positive examples are summed per profile and filtered for locations matching or close to the context.

UAmsCS12wtSUMb

Participants | Proceedings | Input | Appendix

  • Run ID: UAmsCS12wtSUMb
  • Participant: UAmsterdam
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/8/2012
  • Type: both
  • Task: main
  • MD5: 54ff3aef678e28fc081e038f1ca0e93a
  • Run description: Suggestions are derived from Wikitravel, ranked using a standard language model with descriptions of positive examples as queries. Score for suggestions for individual positive examples are summed over all profiles and filtered for locations matching or close to the context.

udelnp

Participants | Input | Appendix

  • Run ID: udelnp
  • Participant: udel
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/10/2012
  • Type: preference
  • Task: main
  • MD5: 2cb21809a5ea2bdeb7628f642cc1743c
  • Run description: In this run we take into account the examples which were liked and example sites which were not liked by the profile. We score the suggestions by calculating the log probability of generating the good examples site and subtracting the log probability of generating the bad example sites from the suggestion site. We then rank the suggestion sites on decreasing score. We consider the suggestion site as multinomial distribution of terms and then take this distribution to calculate the log probability of generating the example site.

udelp

Participants | Input | Appendix

  • Run ID: udelp
  • Participant: udel
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/10/2012
  • Type: preference
  • Task: main
  • MD5: b87c1c532c7d70cc450276deaf82037b
  • Run description: Suggestions only based on examples liked by profile. We use language modeling to score. We consider the suggestion sites as multinomial distribution.

UDInfoCSTc

Participants | Proceedings | Input | Appendix

  • Run ID: UDInfoCSTc
  • Participant: udel_fang
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 7/30/2012
  • Type: both
  • Task: main
  • MD5: e4569967680cb0c32e0d0e12277b9a8d
  • Run description: We use the categories of examples. For each user(profile), all examples are separated into two parts -- positive ones and negative ones, scores of suggestions are computed as alphaavg(pos)-(1-alpha)avg(neg) where alpha is decided by doing cross-validation learning on examples. For context, suggestions are retrieved from yelp and foursquare using and radius. As for time, we collected the business hours on yelp as training data and generate a mapper.

UDInfoCSTdc

Participants | Proceedings | Input | Appendix

  • Run ID: UDInfoCSTdc
  • Participant: udel_fang
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 7/30/2012
  • Type: both
  • Task: main
  • MD5: 959f807baaf9e8f2244a8a59a9d62688
  • Run description: This run uses both categories and descriptions. For each potential suggestion, the score is computed as betadescription_score+(1-beta)categories_scores where beta is got from training examples. And description_score is computed as the following---use description as queries and web pages of potential suggestions as corpus; the retrieval function used is F2EXP of axiomatic method. Categories_score is computed using modified Jaccard Similarity. For each user(profile), all examples are separated into two parts -- positive ones and negative ones, scores of suggestions are computed as alphaavg(pos)-(1-alpha)avg(neg) where alpha is decided by doing cross-validation learning on examples. For context, suggestions are retrieved from yelp and foursquare using and radius. As for time, we collected the business hours on yelp as training data and generate a mapper.

watcs12a

Participants | Proceedings | Input | Appendix

  • Run ID: watcs12a
  • Participant: isi_paik
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: 0455896b992e562334fffb7284e3c00a
  • Run description: To rank suggestions I first sorted the suggestions in the order of final preference, then in the order of initial preference and then the score computed by geo-temporal information.

watcs12b

Participants | Proceedings | Input | Appendix

  • Run ID: watcs12b
  • Participant: isi_paik
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: 9d4066a21cacef9173b0bcb49c278d89
  • Run description: To rank suggestions I first sorted the suggestions in the order of final preference, then in the order of initial preference and then the score computed by geographic information only.

waterloo12a

Participants | Input | Appendix

  • Run ID: waterloo12a
  • Participant: waterloo
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/9/2012
  • Type: both
  • Task: main
  • MD5: 243b7f21071d2c27747a6d3a67654574
  • Run description: Top 50 attractions for each city (and in some cases neighbouring cities) from tripadvisor.com. Ignored profiles.

waterloo12b

Participants | Input | Appendix

  • Run ID: waterloo12b
  • Participant: waterloo
  • Track: Contextual Suggestion
  • Year: 2012
  • Submission: 8/10/2012
  • Type: both
  • Task: main
  • MD5: f11c2e187a7847c8cfc396c3844767a4
  • Run description: Results generated by searching tripadvisor.com using manually generated terms based on examples. Final values in profiles determine which search results are combined to give suggestions.