TREC 2014 Contextual Suggestion Track Guidelines
The Contextual Suggestion Track investigates search techniques for complex information needs that are highly dependent on context and user interests.
Track Organizers:
Charles L A Clarke, University of Waterloo (Primary contact: claclark@plg.uwaterloo.ca)
Adriel Dean-Hall, University of Waterloo
Jaap Kamps, University of Amsterdam
Paul Thomas, CSIRO
Track webpage: https://sites.google.com/site/treccontext/
Mailing list: Send a mail message to listproc (at) nist.gov such that the body consists of the line subscribe trec-context <FirstName> <LastName>
What's New
The Contextual Suggestion Track will be running for a third time for TREC 2014, here are the major differences from last year:
CSV will be the only file format used for profiles, contexts, and returned suggestions.
The number of example suggestions in the profiles will be increased to 100 from 50, and will be from two cities instead of one.
Background
According to a report from the The Second Strategic Workshop on Information Retrieval in Lorne (published in the SIGIR Forum, June 2012): “Future information retrieval systems must anticipate user needs and respond with information appropriate to the current context without the user having to enter an explicit query... In a mobile context such a system might take the form of an app that recommends interesting places and activities based on the user’s location, personal preferences, past history, and environmental factors such as weather and time... In contrast to many traditional recommender systems, these systems must be open domain, ideally able to make suggestion and synthesize information from multiple sources...”
For example, imagine a group of information retrieval researchers with a November evening to spend in beautiful Gaithersburg, Maryland. A contextual suggestion system might recommend a beer at the Dogfish Head Alehouse (www.dogfishalehouse.com), dinner at the Flaming Pit (www.flamingpitrestaurant.com), or even a trip into Washington on the metro to see the National Mall (www.nps.gov/nacc). The goal of the Contextual Suggestion track is to provide a venue for the evaluation of such systems
Task Summary
As input to the task, participants will be given a set of profiles, a set of example suggestions, and a set of contexts. Details of all file formats are given in separate sections below. Each profile corresponds to a single user, and indicates that user’s preference with respect to each example suggestion. For example, one suggestion might be to have a beer at the Dogfish Head Alehouse, and the profile might include a negative preference with respect to this suggestion. Each training suggestion includes a title, description, and an associated URL. Each context corresponds to a particular geographical location (a city). For example, the context might be Gaithersburg, Maryland.
For each profile/context pairing, participants should return a ranked list of up to 50 ranked suggestions. Each suggestion should be appropriate to the profile (based on the user’s preferences) and the context (according to the location). The description of the suggestion may be tailored to reflect the preferences of that user. Profiles correspond to the stated preferences of real individuals, who will return to judge proposed suggestions. Users recruited through crowdsourcing sites or are university undergraduate and graduate students. For the purposes of this experiment, you can assume users are of legal drinking age at the location specified by the context. You may assume that the user has up to five hours available to follow a suggestion and has access to appropriate transportation (e.g., a car).
Timeline
Guidelines finalized: April 28
Contexts available: April 28
Profiles available: June 2 June 9
Submissions due: July 21 July 28
Results available: September
Example Suggestions & Profiles
Profiles consist of two ratings for a series of attractions, one rating for the attraction's title and description and one rating for the attraction's website. The profile should gives systems which attractions a particular user likes and which ones the user does not like. The ratings are given on a five-point scale based on how interested the user would be in going to the attraction if they were visiting the city it was in:
4: Strongly interested
3 Interested
2: Neutral
1: Disinterested
0: Strongly disinterested
-1: Website didn't load or no rating given
Profiles will be distributed as two files:
examples2014.csv: contains the example suggestions that were rated by users
profiles2014.csv: contains the ratings given for the example suggestions
exaples2014.csv is a CSV file with the columns id (attraction_id), title, description, and url, for example:
id,title,description,url
1,Fresh on Bloor,"Our vegan menu boasts an array of exotic starters, multi-layered salads, filling wraps, high protein burgers and
our signature Fresh bowls.",http://www.freshrestaurants.ca
...
profiles2014.csv is a CSV file with the columns id (profile_id), attraction_id, description, and website, for example:
id,attraction_id,description,website
1,1,1,0
1,2,3,4
...
2,1,4,4
...
Contexts
Contexts will be distributed as one file, contexts2014.csv, which contains the cities suggestions need to be generated for. The cities in the contexts file will be the primary cities of 50 randomly selected metropolitan areas (which are not part of a larger metropolitan area) excluding the two seed cities in the example suggestions*.
contexts2014.csv is a CSV file with the columns id (context_id), city, state, lat, and long, for example:
id,city,state,lat,long
1,New York City,NY,40.71427,-74.00597
2,Chicago,IL,41.85003,-87.65005
...
The latitude and longitude and provided as a convenience and is intended to be synonymous with the city and state information.
* The list of metropolitan areas is gathered from http://en.wikipedia.org/wiki/List_of_metropolitan_areas_of_the_United_States
Suggestions
Suggestion files should contain a list of up to 50 ranked suggestion for each profile+context pair are returned as a single CSV file.
Your submitted CSV file should contain the columns groupid, runid, profile, context, rank, title, description, url, and docId, (in that order) for example:
groupid,runid,profile,context,rank,title,description,url,docId
group44,run44A,1,1,1,Deschutes Brewery Portland Public House,"Deschutes Brewery’s distinct Northwest brew pub in Portland’s
Pearl District has become a convivial gathering spot of beer and food lovers since it’s 2008 opening.",
http://www.deschutesbrewery.com,
The suggestions file contains:
groupid: Your official TREC group id
runid: A unique identifier for your run
title: Up to 64 characters long (can be tailored to the user), you may quote this field
description: Up to 512 characters long (can be tailored to the user), you may quote this field
url: a URL that identifies the suggestion on the open web (if docId contains anything this should be empty)
docId: a ClueWeb12 WARC-TREC-ID as described here lemurproject.org/clueweb12.php/specs.php (if url contains anything this should be empty)
Submissions will be identified as either ClueWeb runs of open web runs. Submitted files must consists of suggestions with either only ClueWeb12 docIds or only urls from the open web. Note that we cannot guarantee that judgements for ClueWeb docs will be based off of documents in the ClueWeb12 corpus, docIds may simply be translated to urls and judged off the open web.
The Contextual Suggestion Track will use version 1.1 of ClueWeb12 (which fixes a duplicate document problem in v1.0). Further information regarding the collection can be found on the ClueWeb12 website. Since it can take several weeks to obtain the dataset, we urge you to start this process as soon as you can.
If you are unable to work with the full ClueWeb12 dataset, we will accept runs over the smaller ClueWeb "Category B" dataset (called ClueWeb12-B-13) but we stongly encourage you to use the full dataset if you can. Additionally, both the full dataset and the B-13 dataset will be ranked together during evaluation.
Each group may submit either one or two runs, which should use the same group id but different run ids.
Judging
Suggestions will be judged both by users and NIST assessors:
The user corresponding to each profile will judge suggestions in the same way as the examples, assigning a value of 0-4 for each title/description and url (or ClueWeb doc).
NIST assessors will judge suggestions in term of geographical appropriateness on a three-point scale:
2: good geographical fit
1: marginal geographical fit (distance may be too far)
0: not geographically appropriate
ClueWeb and open web submissions will be ranked separately.
Evaluation Metrics
P@5 (main), MRR, and TBG are the three measures that will be used for the TREC 2014 Contextual Suggestion Track.