Community Help: Discovering Tools and Locating Experts in a Dynamic Environment

Keywords

(*) Unix is a trademark of Novell, Inc.

Introduction

Finding the right person to ask can be time consuming and can interrupt the work of multiple persons. A particular question is only answered to a single person, i.e. another person with the same question has to interrupt again. Furthermore, there is always the risk that a question can reveal incompetence with negative consequences for the person asking the question.

In this paper we report about on-going work on a prototype of a community sensitive help system called ToolBox. ToolBox actively collects information about Unix tools that are used on a collection of networked Sun 4 workstations at the author's address. Information is collected by tapping into accounting information generated by the operating system and by interviewing users that are selected on the basis of collected information. More specifically, ToolBox collects information via three processes: (i) tool discovery and user selection, (ii) user soliciting, and (iii) interviewing. The goal of the tool discovery and user selection process is to discover new tools and locate people that are most knowledgable about a particular tool. The goal of the soliciting process is to convince users to give an interview, and the goal of the interviewing process is to exchange valuable information with the user. The result is a help system that continually seeks to update itself, that contains information that is entirely based on the community's perspective on tools, and that consequently grows with the community and its dynamic environment.

RELATED WORK

Answer Garden [1] relieves the user from asking questions by presenting a branching network of diagnostic questions from which the user can choose the best match to his/her problem. If no satisfactory match exists the user can press an "Unhappy" button and post the new question which is automatically routed to the appropriate expert. The system is explicitly configured in order to provide this routing service, i.e., experts are known a priori. In the advent of new tools, Answer Garden does not support the discovery of new experts. The authors assume formally assigned experts whose job is to maintain Answer Garden. DYK ("Did You Know") [7] allows any member of a community to add new pieces of information to a directory service with pre-defined categories. The idea is to disseminate information that people consider important and which is hard to find in documentation. DYK randomly presents these pieces of information to users who invoke the tool and select one of the categories. In order to motivate this information exchange the author designed DYK such that users like to use it as some form of procrastination. ToolBox does not rely on randomly collecting and disseminating information. Instead, it interviews selected users who construct and modify an information space according to their community specific understanding. ToolBox provides a very structured and customizable information space that helps users to retrieve information in the situation of a particular problem. Information exchange is motivated by providing occasions for interviews.

TOOLBOX

The WWW server (e.g., httpd) is extended by programs that dynamically generate Hypertext Markup Language (HTML) documents. The Picker process discovers new tools and locates their users by periodically scanning and analyzing process tables on a collection of hosts. The Solicitor process sends email interview invitations to selected users. The goal of the user selection process is to find users that are knowledgable and most likely to give interviews. The goal of the interview invitation process is to make it easy for the user to start an interview and convey a clear and very compact scope of the suggested interview to the user. The Newspaper process periodically advertises the ToolBox services to the community by sending out weekly email newsletters which (i) contain statistics about tool usage, (ii) acknowledge people who gave interviews to the ToolBox, (iii) announce newly discovered tools, and (iv) give general information about ToolBox.

The client can be any WWW client that supports HTML form processing. Together with the extended WWW server the client serves either as a viewer of the ToolBox information space or as the front end of the Interviewer by guiding a user through a set of interactive hypertext documents. These documents allow the user (i) to freely restructure the categorization of tools by moving tools and categories and by adding new categories, (ii) to categorize new tools, and (iii) to add other information that s/he considers relevant. The Interviewer facilitates a goal directed exploration of the ToolBox information space, the goal being the categorization of a small collection of tools. We assume that during this process the user is likely to discover new tools that are relevant to his/her work. The argument is: since the user tries to categorize a collection of tools s/he frequently uses, s/he is likely to explore the part of the information space that seems to be most related to this tool collection and, thus, to his/her work. Hence, the ToolBox and the user are exchanging information during the interview

STATUS AND FUTURE RESEARCH

Although every community member could freely modify the information space of ToolBox it stabilized quickly after a few days of intense restructuring and growth: 63 categories were added within the first few days. At the end of the month only six more categories were added. The categories were redundant in only two cases.

Work on improving the ToolBox prototype continues. Users have provided very positive feedback about the ToolBox concept. On-going research consists of (i) finding new ways to improve the user selection strategy of the Solicitor, (ii) improving the interviewing process, (iii) exploring feedback mechanisms in order to distinguish the quality of contributions, and (iv) validating the ToolBox concept by measuring the average number of tools known to a community member before and after a test run.