Logged conversation must be conceived of as an OM (Organizational Memory) manufacturing process acquired during active discussions among conversation participants and logged into a hierarchy of text-format data. A mailing list provides a collection of informative correspondences, and its use as a source of organizational knowledge in business is widespread and obvious. To make this use easier, however, good search tools are required. The plethora of current and widespread search technologies can be classified into four categories:
(1) Context-extended search using a thesaurus. In this approach, the query term is expanded in context utilizing words/clauses/phrases from a thesaurus to broaden the catchment space.
(2) Query-by-example. Here, the user selects relevant document snippets, which are then used for the query base.
(3) Keyword Search on full/partial text contents.
4) List Scan, by which the user sorts by date, author, and subject by thread listing within the structured list (See Figure 1).
[FIGURE 1 OMITTED]
Experimental Setting and Results
From the collection of 20 mailing lists gathered for this study, we found on analysis that they consist of three primary types:
1. In "strongly-typed" mailing lists, each reply or thread-initiating message can only start after the completion of a "category" input. A typical example of this type of mailing list is Organizational Memory Information Systems (OMIS), which force users to input a category type into the system, often presenting an unnecessary entry barrier. Such "category type" information can significantly enhance later OMIS search usability and effectiveness, but our observation has been that the threads do not extend to more than 10 messages. Here, we suspect the negative effect of the "entry barrier" outweighs usability, and we cannot argue that this type of mailing list is good for our thread classification (Mark & Bordetsky, 1998).
2. A typical "expert locating" system allows each member to register and modify the index list pertaining to his or her specialized knowledge or experiences so that later users can locate subject matter experts by a simple keyword query. Each expert consultant will be registered, and the collected expert profiles form a "thread" in the mailing list. Users can add comments and feedback, rating the services received. Such registration data also constitutes a mailing list. It is rare here, as in type (1), to see sufficiently long threads. Another drawback is that OM manufacturing does not happen in the "expert locating" system per se. Usually, the OM is created while messages are exchanged in a "weakly typed" message swap (Fisher, 2002). For these reasons, it is inadvisable to choose the "expert locating" system for thread analysis and classification.
3. The "plain-vanilla free format" or "weakly-type-forced" mailing lists is often used for hobby and special interest groups. In enterprise, employees form "intra-net" mailing list to assist OM creation and utilization (Hood, 2003).
The mailing list we selected for analysis was borrowed from the amateur radio kit building community (Elecraft, 2003), where OM processes are actively performed and databases frequently accessed. Our list displays distinctive aspects arising from the nature of this community:
1. The list has been used since 1998 and the total number of messages per month increased constantly from the start.
2. The January 2003 list consists of 1406 messages by 311 members. As shown in Figure 2, 156 authors (50%) account for 86% of the messages, forming the list's active core. Fortyseven authors (15%) account for 54% of the messages.
3. Visitors and infrequent guests welcomed by "expert" advisers constitute 63% of the posters, posting only once or twice in the course of a typical month.
Because of these well-behaved community aspects, OM created in the above list qualified as a relevant analysis corpus for our thread classification. …