�ݺ�ߣ

Inferring User Tasks and Needs
Rishabh Mehrotra1, Emine Yilmaz2, Ahmed Hassan Awadallah3
1Spotify, London
2University College London
3Microsoft Research

Outline of the Tutorial
• Section 1: Introduction
• Section 2: Characterizing Tasks
• Section 3: Tasks Extraction Algorithms
• Section 4: Task based Evaluation
• Section 5: Applications

Section 5: Applications
– Task based personalization
–Task based recommendations
–Task Tours
–Predicting Task Continuation
–Task Completion Dialogue Systems

Task Based Personalization
• Users tend to be interested in certain tasks when they use
search engines
• Represent users in terms of tasks they are interested in
• Personalize search results based on that
– Query recommendation, re-ranking search results, …

Traditional Approach: Topic Based Personalization
• Topics commonly
constructed in two ways
– Manual topical
categories (e.g. ODP)
– Topic Modelling Based
(e.g. LDA)
Topics

Personalization: Topics versus Tasks
Topics Tasks
Finance,
Basketball, Jazz
Finance,
Basketball, Pop
music
Basketball, Pop
music

Task based Personalization
[Mehrotra and Yilmaz, ACM RecSys Posters ’14]
– Given N users and M tasks
– Construct an NxM user-task association matrix R
• Cosine similarity between user profiles and task representations
– Find the Nxd user feature matrix U, and Mxd task feature matrix
T s.t.
R ≈ U TT
– Discover the d latent features underlying the interactions
between users and tasks

Representing Users in the Task-space
• Existing user modeling methods fail to differentiate between
users having similar topical interests
– User curious about "search engines" and an experienced IR researcher
– a stockbroker and a normal investor
• The objective is to leverage user's topical interest profiles
along with user's task associations.
Topics Tasks
Finance,
Basketball, Jazz
Finance,
Basketball, Pop
music
Basketball, Pop
music

Coupling Topics and Tasks
[Mehrotra and Yilmaz, ACM ICTIR’15]

• Construct a 3-mode tensor to jointly model the user's topical
and task preferences:
– <users, topics, tasks>
• Define each tensor component as:
• A user's participation in a certain task gets weighted by her
topical affinity.

• Tensor decomposition to leverage connections between different
users across different topics and different tasks
• PARAFAC Tensor Decomposition [Stegeman and Sidiropolous, Linear Algebra and
Applications, ‘07]
• Ui, Vj, Tk are D-dimensional vectors representing users, topics, and
tasks, respectively
• Discover the D latent features underlying the interactions between users, topics and tasks

Evaluation: Collaborative Query Recommendation
• Identify user cohorts based on user
preferences
• Personalize search results based on
recommendations from similar users
Number of Similar Users

• Provide heterogeneous recommendations during users’
browsing process
• Define tasks as demand sequences embedded in user browsing
sessions
Task-based Recommendation on a Web-Scale

• Step 1: Collaborative Task Mining:
– extract frequent demand sequences from large scale browser logs
– achieved via frequent sequence mining problem
• Step 2: Task-based Demand Prediction
– predict the upcoming demand of a user given the current browsing session
– estimate the probability of each demand d ∈ D being the follow-on demand of
the current session
• Step 3: Task-based Recommendation
– Provide site-level recommendations (based on predicted demands)
– Provide link-level recommendations (heterogeneous recommendations
based on browsing behavior)
Task-based Recommendation on a Web-Scale

Task Tours: Helping Users Tackle Complex Search Tasks
Automatically create multi-step task tours:
– URL labeling with topical category to identify tasks
– construction of the task graph that relates tasks to
each other
– building of the tours using task graph
– identification of triggers
Task tours help users:
– understand the required steps to complete a task,
– find URLs related to the active task
– alert users to activities they may have missed
Task Tours: Helping Users Tackle Complex Search Tasks, CIKM 2012

Predicting Task Continuation
– Understand, characterize and
detect tasks which will be
continued
– Bing logs used to identify intent,
topics & search behavior
associated with long running
tasks
– Prediction model using various
features
Search, Interrupted: Understanding and Predicting Search Task Continuation; SIGIR 2012
Task continuation for broad search intent

Task Completion Dialogue Systems
– Reinforcement learning based
model
– Goal directed conversations
– Accesses external knowledge
base
– Slot filling to form a semantic
frame
End-to-End Task-Completion Neural Dialogue Systems, arXiv 2017

Summary - I
– Query intent understanding
• Classification based (ODP, LDA)
• Cluster based (Random walks, reformulations)
• Session based techniques
– Time based segmentation
– Content based segmentation
– Hybrid segmentation
– Extracting search tasks
– Evaluating task extraction algorithms
– Applications

• Task Extraction
– Clustering based approaches
– Entity oriented task extraction
– Structured SVM based bestlinks structures
– LDA topics with Hawkes process
• Tasks à Subtasks
– dd-CRP with embeddings model
– BRT Hierarchical Subtask segmentation
– Applications
Summary - II

• Gold standard dataset
• User study based evaluation
• Alternative techniques
• TREC Tasks Tracks
– Applications
Summary - III

–Query intent understanding
–Extracting search tasks
–Evaluating task extraction algorithms
–Applications
• Task based user modeling
• Related Search suggestions
• Task based ecommerce recommendations
Summary - IV

Ongoing/Future Work
• Task based user satisfaction prediction
• Digital assistants
– Task understanding
– Task completion
• Book Uber
• Deliver food
• Task based recommendations
• Beyond search – web tasks

Questions?
• Rishabh Mehrotra
Research Scientist
Spotify, London
rishabhm@spotify.com
• Emine Yilmaz
Associate Professor, UCL
Faculty Fellow, The Alan Turing Institute
Research Consultant, Microsoft Research
emine.yilmaz@ucl.ac.uk
• Ahmed Hassan Awadallah
Research Lead
Microsoft Research, Redmond
hassanam@microsoft.com

References
[1] Ahmed, White, Pantel, Dumais, and Wang. Supporting complex search
tasks. In Proceedings of the ACM CIKM 2014.
[2] Baeza-Yates, Hurtado, and Mendoza. Query recommendation using query
logs in search engines. In Current Trends in
Database Technology-EDBT 2004 Workshops, 2005.
[3] D. M. Blei and Griths. The nested chinese restaurant process and bayesian
nonparametric inference of topic hierarchies.
Journal of the ACM (JACM) 2010.
[4] Blundell and Teh. Bayesian hierarchical community discovery. In NIPS 2013.
[5] C. Blundell, Y. W. Teh, and K. A. Heller. Bayesian rose trees. In UAI 2010.
[6] B. Carterette, E. Kanoulas, M. Hall, and P. Clough. Overview of the trec 2014
session track. 2013.
[7] L. D. Catledge. Characterizing browsing strategies in the world-wide web.
Computer Networks and ISDN systems, 1995.
[8] Chuang and Chien. Towards automatic generation of query taxonomy: A
hierarchical clustering approach. In ICDM 2003.
[9] N. Craswell and M. Szummer. Random walks on the click graph. In ACM
SIGIR 2007.
[10] Donato, Bonchi, and Chi. Do you want to take notes?: identifying missions
in yahoo! search pad. In WWW 2010.

[11] D. He. Combining evidence for automatic web session identication.
Information Processing & Management, 2002.
[12] Heller and Ghahramani. Bayesian hierarchical clustering. In ICML 2005.
[13] Hua, Song, and Wang. Identifying users' topical tasks in web search. In
ACM WSDM 2013.
[14] Jones, Rey, Madani, and Greiner. Generating query substitutions. In
WWW 2006.
[15] R. Jones and K. L. Klinkner. Beyond the session timeout: automatic
hierarchical segmentation of search topics in query logs. In CIKM 2008.
[16] J. H. Kim and Kim. Modeling topic hierarchies with the recursive
chinese restaurant process. In CIKM 2012.
[17] Kotov, Bennett, White, Dumais, and Teevan. Modeling and analysis of
cross-session search tasks. In SIGIR 2011.
[18] L. Li and H. Deng. Identifying and labeling search tasks via query-based
hawkes processes. In KDD 2014.
[19] Y. Li. Relationships among work tasks, search tasks, and interactive
information searching behavior. ProQuest, 2008.
[20] Y. Li and N. J. Belkin. A faceted approach to conceptualizing tasks in
information seeking. Information Processing & Management, 2008.
References

[21] H. Liao, Song. Evaluating the effectiveness of search task trails. In WWW
2012.
[22] J. Liu and N. J. Belkin. Personalizing information retrieval for multi-
session tasks: The roles of task stage and task type. In SIGIR 2010.
[23] Lucchese, Orlando, Perego, Silvestri, and Tolomei. Discovering tasks from
search engine query logs. ACM Transactions on Information Systems, 2013.
[24] C. Lucchese, S. Orlando, R. Perego, F. Silvestri, and G. Tolomei. Identifying
task-based sessions in search engine query logs. In WSDM 2011.
[25] R. Mehrotra, P. Bhattacharya, and E. Yilmaz. Characterizing users' multi-
tasking behavior in web search. In CHIIR 2016.
[26] Mei, Zhou, and Church. Query suggestion using hitting time. In ACM
CIKM 2008.
[27] Q. Mei, H. Fang, and C. Zhai. A study of poisson query generation model
for information retrieval. In SIGIR 2007.
[28] D. Morris. Searchbar: a search-centric web history for task resumption
and information re-nding. In CHI 2008.
[29] D. Newman, J. H. Lau, K. Grieser, and T. Baldwin. Automatic evaluation of
topic coherence. In NAACL 2010.
[30] O'Connor, Krieger, and Ahn. Tweetmotif: Exploratory search and topic
summarization for twitter. In ICWSM 2010.
References

[31] P. Pecina. Lexical association measures and collocation extraction.
Language resources and evaluation, 2010.
[32] F. Radlinski and T. Joachims. Query chains: learning to rank from implicit
feedback. In KDD 2005.
[33] Segal, Koller, and Ormoneit. Probabilistic abstraction hierarchies. NIPS
2002.
[34] Silverstein and Marais. Analysis of a very large web search engine query
log. In SIGIR Forum 1999.
[35] A. Singla, R. White, and J. Huang. Studying trailnding algorithms for
enhanced web search. In SIGIR 2010.
[36] Song, Liu, and Wang. Automatic taxonomy construction from keywords.
In Proceedings of the 18th ACM SIGKDD 2012.
[37] Spink, Koshman, Park, Field, and Jansen. Multitasking web search on
vivisimo. com. In ITCC 2005.
[38] P. Vakkari. Task-based information searching. Annual review of
information science and technology, 2003.
[39] H. Wang, X. Song, R. W. White, and W. Chu. Learning to extract cross-
session search tasks. In WWW 2013.
[40] White, Bennett, and Dumais. Predicting short-term interests using
activity-based search context. In CIKM 2010.
References

•Deadline: 30th November 2017
•Notification: 15th December 2017
•Workshop: 9th February 2018
aka.ms/wsdm2018-learnir-workshop

�ݺ�ߣ

Parts 5 & 6: WWW 2018 tutorial on Understanding User Needs & Tasks

Recommended

More Related Content

Similar to Parts 5 & 6: WWW 2018 tutorial on Understanding User Needs & Tasks (20)

Recently uploaded (20)

Parts 5 & 6: WWW 2018 tutorial on Understanding User Needs & Tasks