=Paper=
{{Paper
|id=Vol-532/paper-15
|storemode=property
|title=Cobot: Real Time Multi User Conversational Search and Recommendations
|pdfUrl=https://ceur-ws.org/Vol-532/paper15.pdf
|volume=Vol-532
}}
==Cobot: Real Time Multi User Conversational Search and Recommendations==
https://ceur-ws.org/Vol-532/paper15.pdf
Cobot: Real Time Multi User Conversational Search and
Recommendations
Saurav Sahay Anushree Venkatesh Ashwin Ram
College of Computing College of Computing College of Computing
Georgia Tech Georgia Tech Georgia Tech
ssahay@cc.gatech.edu avenkatesh6@gatech.edu ashwin@cc.gatech.edu
ABSTRACT
Cobot is a new intelligent agent platform that connects users This paper introduces a novel Conversational Search and
through real-time and off-line conversations about their health Recommendation system that involves finding relevant in-
and medical issues. Intelligent web based information agents formation based on social interactions and feedback along
(conversational/community bots) participate in each conver- with augmented agent based recommendations. People in
sation providing highly-relevant real-time informational rec- social groups can provide solutions (answers to questions)[3],
ommendations and connecting people with relevant conver- pointers to databases or other people (meta-knowledge)[3][6],
sations and other community members. Cobot provides an validation and legitimation of ideas[3][4], can serve as mem-
innovative approach to facilitate easier information access al- ory aids[7] and help with problem reformulation[3]. “Guided
lowing users to exchange information through a natural lan- participation”[11] is a process in which people co-construct
guage conversational approach. Conversational Search(CS) knowledge in concert with peers in their community[12]. In-
is an interactive and collaborative information finding in- formation seeking is mostly a solitary activity on the web
teraction. The participants in this interaction engage in today. Some recent work on collaborative search reports
social conversations aided with an intelligent information several interesting findings and the potential of this technol-
agent (Cobot) that provides contextually relevant factual, ogy for better information access.[5][2][1][9]
web search and social search recommendations. Cobot aims We are building a system called Cobot1 to address some
to help users make faster and more informed search and dis- of these challenges. Cobot introduces a conversational envi-
covery. It also helps the agent learn about conversations ronment that provides social search through conversations
with interactions and social feedback to make better recom- integrated with intelligent semantic meta-search from the
mendations. Cobot leverages the social discovery process by web. Users want to simplify their experience when perform-
integrating web information retrieval along with the social ing an information finding task. Conversational Search is
interactions and recommendations. about letting users collaboratively search and find in natu-
ral language, leaving the task of user intent comprehension
on the system. The participating agent interacts with users
Categories and Subject Descriptors proving recommendations that the users can accept, reject,
H.5.0 [Information Systems Applications]: General; H.3.3 like, dislike or suggest.
[Information Storage and Retrieval]: Information Search
and Retrieval; I.2.7 [Artificial Intelligence]: Natural Lan- 2. SYSTEM DESCRIPTION
guage Processing
Cobot is an intelligent agent platform that connects users
through real-time and offline conversations. Cobot lives
General Terms in a community, has a limited understanding of domains
Design, Human Factors, Algorithms through ontologies and brings relevant information to the
users by participating in the conversations. Cobot’s ’conver-
Keywords sation engine’ monitors user conversations with other users
in the community and provides/receives recommendations
Real time Collaborative Information Access, Social Search, (links and snippets) based on the conversation to the par-
Contextual Collaborative Filtering, Conversational Search ticipants. Cobot’s ‘community engine’ models conversations
to capture user-user and user-information interactions.
1. INTRODUCTION Design Goals
1. Near real time conversational agent
2. Personalized as well as generic recommendations
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are 3. Agent learns with interaction
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, to 4. Uses a structured internal organization of content
republish, to post on servers or to redistribute to lists, requires prior specific 1
permission and/or a fee. We use the term Cobot for Cobot system as well as Cobot
Copyright 200X ACM X-XXXXX-XX-X/XX/XX ...$10.00. agent interchangeably
5. Dynamically connects conversations to the right set of 2.3 Aggregated Web Search
people for participation Identifying relevant documents for a particular user’s need
without extensive search, in conversational manner is the
6. Helps the user talk about health issues. (Real time key objective for precise search. The right search queries
conversations) need to be figured out with situation assessment from the
conversational snippets. It is not desirable to return dozens
7. In the real time conversational process, provides rec- or hundreds of remotely relevant results, even if some of
ommendations. (‘who to talk to’, ‘what to look at’) them will be highly relevant. The aim is to retrieve succes-
sive recommendations that try to address the search prob-
In the following sections, we briefly describe the features
lem precisely. Cobot uses different shallow semantic parsing
of the Cobot system.
techniques for operationalizing a user’s intent into computa-
2.1 Real time Conversational Search tional form, dispatching to multiple, heterogeneous services,
gathering and integrating results, and presenting them back
Cobot provides a conversational interface that combines to the user as a set of solutions to their request.
semantic language understanding and real time collabora-
tive environment for information retrieval with contextually 2.4 Real time matching of participants to con-
relevant recommendations. Cobot helps people find infor- versations
mation faster with the aim for finding useful responses for
Communities are made up of users who are grouped by
completing the search intent. The conversational interface
different information needs into dynamic cohorts. These on-
allows for much more interactivity than one-shot search style
line communities, through effective sharing and collabora-
interfaces, which aids usability and improves intent under-
tion, increase the utility of systems and help solve individ-
standing. For example, Cobot recommends links and snip-
ual problems more effectively. Cobot allows for connecting
pets from relevant articles on the internet. It also makes so-
two or more individuals to an online conversation based on
cial recommendations to connect contextually relevant users
the topic and context of conversation, mutual interests, and
to the conversation.
what they want to talk about at that time. The system
The approach we have taken to address CS problems is
allows any individual to find/join that conversation.
by developing dynamic data structures that model it. We
call this structure the “Socio-Semantic Model” - these con-
versation nets maintain in memory models of the conver-
2.5 Socio-Semantic Collaborative Filtering
sation, participants, participants’ immediate social connec- Filtering and recommendation are crucial in collaborative
tions, concepts, relationships and information flow. systems enabling users to navigate an ever-growing deluge
of information more effectively. Cobot’s recommendation
2.2 Socio-Semantic Model engine delivers quality information delivered through filters
The Socio-Semantic Conversation Model is a dynamic mem- achieved from semantic and contextual understanding of
ory data structure based on principles of experience based text along with captured users’ interests. It uses various
agent architecture.[10] It supports interleaved retrieval of personalization techniques such as collaborative filtering on
information by applying different memory retrieval algo- conversations and other entities in context. Natural lan-
rithms. The model maintains the user’s social graph, the guage processing techniques are used to enhance the content
conversation graph with the extracted semantic net for the based recommendations.[8]
conversation.
Some essential properties of the model are as follows: 3. SYSTEM ARCHITECTURE
Figure 2 depicts the high level architecture of the Cobot
• The model is socially aware of the participant and system. The Conversational Agent uses different modules
his social network’s availability (to aid with Cohort for conversation analysis, search and recommendation and
Matching) maintains a short-term conversation memory for each con-
versation. The socio-semantic model/net is analogous to
• The model provides bi-directional recommendation and
the agent’s long term memory model where it stores all pro-
feedback. (Both agent and the participant can add rec-
cessed information about users, conversations, activities and
ommendations)
content descriptors.
• The model understands limited domain terminology
and is able to find semantic relationships amongst con- 4. SYSTEM PROTOTYPE
cepts extracted from conversations. Figure 3 shows one screenshot of the initial system proto-
type which is work in progress. This prototype is designed
• The model is aware of user’s profile (such as interests for health related searches by incorporating medical ontolo-
and ratings) for the agent to be able to use that infor- gies. Users actively engage in conversations by multi-user
mation. chat, rating or adding recommendations. The agent moni-
tors the environment to build user interaction models and
The Socio-Semantic Model aims to provide storage and to improve search relevance.
memory based retrieval for dynamic representation, update
and reuse of users’ knowledge and experiences. Figure 1
depicts the user-centric domain information modeling ap- 5. CONCLUSION
proach to jointly model the information context from users’ This paper proposes a collaborative system for conversa-
perspective. tional search and recommendations. We are hypothesizing
Figure 1: Socio-Semantic Net
Figure 2: System Architecture
Figure 3: Prototype Interface
that such a Conversational Search system is more usable international conference on Intelligent user interfaces,
for information access as compared to a solitary web search pages 171–177. ACM New York, NY, USA, 2006.
experience. We briefly describe the design goals and fea- [6] E. A. Fox, D. Hix, L. T. Nowell, D. J. Brueni, D. Rao,
tures involved in construction of the Cobot system. Socio- W. C. Wake, and L. S. Heath. Users, user interfaces,
and objects: Envision, a digital library. J. Am. Soc.
Semantic Conversation Modeling using Experience-based Agency Inf. Sci., 44(8):480–491, 1993.
is a unified approach for addressing Conversational Search [7] I. Karasavvidis. Distributed Cognition and
problem. The dynamic and self configuring memory struc- Educational Practice. Journal of Interactive Learning
tures and the semantic net details enable memory retrieval Research, pages 11–29, 2002.
from the storage. Automatic Cohort Matching based on [8] P. Melville, R. Mooney, and R. Nagarajan.
Conversations and User Profiles incorporate a methodology Content-boosted collaborative filtering for improved
to dynamically pull users for conversations. Unlike users recommendations. In Proceedings of the National
Conference on Artificial Intelligence, pages 187–192.
themselves having to find relevant conversations, the con- Menlo Park, CA; Cambridge, MA; London; AAAI
versations find the users using this approach. Press; MIT Press; 1999, 2002.
[9] S. Paul and M. Morris. Cosense: enhancing
sensemaking for collaborative web search. In
6. REFERENCES Proceedings of the 27th international conference on
[1] S. Amershi and M. R. Morris. Cosearch: a system for Human factors in computing systems, pages
co-located collaborative web search. In CHI ’08: 1771–1780. ACM New York, NY, USA, 2009.
Proceeding of the twenty-sixth annual SIGCHI [10] A. Ram and A. Francis. Multi-plan retrieval and
conference on Human factors in computing systems, adaptation in an experience-based agent. Case-Based
pages 1647–1656, New York, NY, USA, 2008. ACM. Reasoning: experiences, lessons, and future directions,
[2] O. Boydell and B. Smyth. Enhancing case-based, pages 167–184, 1996.
collaborative web search. Lecture Notes in Computer [11] B. Rogoff. Apprenticeship in thinking: Cognitive
Science, 4626:329, 2007. development in social context. Oxford University Press
[3] R. Cross, R. E. Rice, and A. Parker. Information New York, 1990.
seeking in social context: structural influences and [12] B. Wilson and H. Meij. Constructivist learning
receipt of information benefits. IEEE Transactions on environments: Case studies in instructional design.
Systems, Man, and Cybernetics, Part C, IEEE Transactions on Professional Communication,
31(4):438–448, 2001. pages 0361–1434, 1997.
[4] B. M. Evans and E. H. Chi. Towards a model of [13] O. Ybarra, E. Burnstein, P. Winkielman, M. C.
understanding social search. In CSCW ’08: Keller, M. Manis, E. Chan, and J. Rodriguez. Mental
Proceedings of the ACM 2008 conference on Computer Exercising Through Simple Socializing: Social
supported cooperative work, pages 485–494, New York, Interaction Promotes General Cognitive Functioning.
NY, USA, 2008. ACM. Pers Soc Psychol Bull, 34(2):248–259, 2008.
[5] D. Feng, E. Shaw, J. Kim, and E. Hovy. An intelligent
discussion-bot for answering student queries in
threaded discussions. In Proceedings of the 11th