Task Recommendation

Date: March 22, 2019
Venue: Stavanger, Norway. Symposium at the IAI group

Please cite, link to or credit this presentation when using it or part of it in your work.

#InformationRetrieval #IR #TaskBasedSearch #TaskCompletionEngines

Transcript

1. Task Recommendation Darío Garigliotti University of Stavanger March 22, 2019
2. An Example
3. An Example Planning a trip to Oslo
4. An Example
5. An Example things to do in oslo
6. An Example things to do in oslo oslo parks
7. An Example things to do in oslo oslo parks viking museum
8. An Example things to do in oslo oslo parks viking museum psg united
9. An Example things to do in oslo oslo parks viking museum psg united dnb
10. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo
11. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo
12. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo Session 1
13. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo Session 1 Session 2
14. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo Session 1 Session 2
15. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo Session 1 Session 2
16. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo Session 1 Session 2
17. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo Missions: M1, M2, M3 Session 1 Session 2
18. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo Missions: M1, M2, M3 Task T1 Session 1 Session 2
19. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo Missions: M1, M2, M3 Task T2 Task T1 Session 1 Session 2
20. An Example things to do in oslo oslo parks viking museum psg united dnb hotels in oslo Missions: M1, M2, M3 Task T3 Task T2 Task T1 Session 1 Session 2
21. Task-based Search Procedural mission
22. Task-based Search Procedural mission Procedural task:
23. Task-based Search Procedural mission Procedural task:
24. Task-based Search Procedural mission Procedural task:
25. Task-based Search Procedural mission Procedural task: A task that can be accomplished by following a sequence of concrete actions or subtasks
26. Task-based Search Procedural mission Procedural task: A task that can be accomplished by following a sequence of concrete actions or subtasks
27. Task-based Search Procedural mission Procedural task: A task that can be accomplished by following a sequence of concrete actions or subtasks
28. Task-based Search Procedural mission Procedural task: A task that can be accomplished by following a sequence of concrete actions or subtasks 1 Choose your location. 2 Choose when you want to go. 3 Plan a loose itinerary. ...
29. Problems • Problem I: single-query task recommendation • Input: a single query (from a procedural mission) • Output: a task ranking • Problem II: mission-aware task recommendation • Input: a procedural mission • Output: a task ranking
30. Methods • Problem I: single-query task recommendation • Text-based document ranking • Learning-to-Rank • Problem II: mission-aware task recommendation • Aggregation of rankings per query in mission, using method for Problem I
31. Research Questions • RQ1: Which one is a solid text-based baseline, in terms of scoring model and document field? • RQ2: How effective is a Learning-to-rank method Vs. the baseline? • RQ3: What are the most useful features when learning to rank tasks? • RQ4: How effective is recommending tasks for a mission Vs. only a single query in the mission?
32. Dataset Construction • 0. Mission detection • Dataset from task-based search literature • 3,873 queries in 1,378 search missions • 1. Identification of procedural missions • Binary experts annotation: "Is the task corresponding to this mission procedural?" • 54 procedural missions, 19 of which are single-query • 2. Relevance judgments • Graded crowd annotation: • "Is this document relevant for the task behind this query?" • Pool of documents from several text-based baselines