BoB Project
BoB, a chatterbot for the University Library
Objectives
The
University Library and the KRDB
KRDB Research Center at the Faculty of Computer Science, Free University Bozen-Bolzano, closely work together in the research and development of new and future-oriented services for the digital information age.
While all users profit from the synergies and results of these activities, at the same time a foundation for further cooperation between the Library and research departments is built.
Background
The number of users of the Library of the Free University of Bozen-Bolzano is increasing yearly, as well as the size of the catalogues and the complexity of Library structure and organization. The catalogues are searched by students as well as professors, research staff and citizens of Bolzano and the neighborhood, with different interests, expertise, and requests. Moreover, the users are of different nationalities and formulate their requests either in German, Italian, or English.
Therefore, the number of questions librarians have to answer daily is increasing enormously. Answering these questions and maintaining the Frequently Asked Questions page in a multilingual format is a time consuming activity that so far has been carried out only by the Library staff.
Therefore, there is the need of automating the answering process as much as possible.
The development of a multi-lingual user friendly automatic dialogue system, that can handle autonomously and without intervention of library staff the majority of user requests, is seen as cost-effective solution to this problem and represents a long term investment for our Library.
Indeed, such an approach has already been experimented with success in similar contexts. For example, "
Stella", is such an automatic dialogue system (also called chatterbot) built at the University of Hamburg, which can handle up to 95% of the user requests regarding the university library.
Structure
The project consists of three phases. The first phase of the project resulted in the implementation of a first fully functional version of a web-based chatterbot application for Frequently Asked Questions (FAQs). The chatterbot provides Library users an enhanced and innovative system able to assist them 24hrs per day in any of the three languages of FUB, English, Italian, and German.
In a second phase, user interaction logs gathered during run-time of the chatterbot will be used for dynamically adapting the statistical models incorporated in the application. On the one hand, adaptation will focus on the individual user, capturing individual language preferences or preferred human-computer interaction styles (e.g., user initiative vs. system initiative dialogue strategies). On the other hand, overall adaptation involves taking advantage of use data as a whole to enhance natural language processing using statistical methods: by automatically improving a rule-based system and learning from the data, the chatterbot becomes gradually more robust.
In the final phase, natural language understanding will be improved at various stages by learning from user interactions, allowing the chatterbot to go beyond the limitations of purely rule-based/ knowledge driven processing (where the system designer tries to anticipate and hard-code any possible user utterance/action).
The resulting software application will go beyond the state of the art of web-based chatterbot clients and will allow users of the University library to access all relevant information via individualized and highly natural conversational interaction.
How BoB answers Questions
On the one hand, BoB stands in the tradition of
chatterbots: it analyzes each user question based on relatively simple pattern matching techniques. Each question that BoB understands is matched by a so-called question pattern, which we have associated with a corresponding system answer.
This technique of formulating question patterns can become rather complicated, considering the many ways a user might ask for essentially the same information (e.g., "What are the opening times?", "When do you close?", "Can I come on Sunday evenings?"), but it still covers a lot of ground.
After certain user input, BoB takes the "initiative" in the dialogue and asks a clarification question, trying to get a more precise idea of what the user wants to find out about.
BoB already goes beyond other chatterbots by using a
machine learning technique to automatically choose the best answer to a user query, given evidence from previously collected dialogues.
Scientifically interesting Issues concerning the Project
From a scientific point of view, BoB is relevant for Interactive Question Answering research in several ways:
- Collecting real human-machine dialogues, which can then be analyzed on different levels, e.g., to study what users ask about, how they ask, and how accurately BoB is able to answer
- Testing models of dialogue context, which will enable BoB to answer naturally occurring, context-dependent follow-up questions in a more accurate way
- Studying new issues arising with BoB's multilinguality: comparing dialogues in BoB's three languages
Here and
here are lists of scientific publications related to BoB and Interactive Question Answering.
Availability of software, applications and test data
Any person or group with scientific interest in the BoB project may access the software and applications created. For further information, please contact
Elisabeth Frasnelli.
Contact & project members
Please contact kirschner _at_ inf.unibz.it in case of problems, questions or suggestions regarding BoB.
Project members
Armin Barducci
Raffaella Bernardi
Francesca Bonin
Matthias Einbrodt
Daniele Gobetti
Manuel Kirschner
Marlene Klug
Ulrike Kugler
Beatrix Lehner
Werner Mores
Carolin Renkewitz
Dajana Ross
Giorgio Stefanoni
Evgeny Thomsen