8.4 User feedback
8.4.2 Explicit user feedback
The idea behind explicit user feedback is asking the user explicitly how useful he/she found a particular resource returned by the system. Obviously, this technique requires that the user actively rate each resource found by the search system. These ratings then form the basis for the user agent to revise and refine its beliefs about the interests of its users.
Users are asked to rate web resources after they have been visited by the user. Usually users are asked to rate these resources according to some predefined scale (e.g. either good, intermediate or completely irrelevant). The user profile is then updated with these ratings, thereby refining the agent’s model of its user.
One of the advantages of using explicit feedback is that it is quite robust. The agent can collect feedback with varying degrees of relevance, from very positive to very negative. This aids in modelling users more precisely.
Another advantage of explicit feedback is that it could be used as the basis for collecting and storing the perceptions of a community of users about resources on the web. This idea is further discussed in the following chapter.
The main disadvantage of using an explicit feedback scheme is that the user must be willing to rate resources after he/she has considered it. Many users may feel that the rating of resources in this way is unnecessarily time consuming and may not be willing to spend time to assist a user agent with the refinement of his/her user profile. Studies have revealed however, that some users may be willing to spend more time with search systems in order to gain more accurate results [42]. Using ratings as an explicit feedback is the primary feedback mechanism of many personalized search systems [66, 85].
Implicit and explicit feedback can both be used to refine the user agent’s perceptions about its user’s interests. In the context of the user modelling profile presented in this section, this ef-fectively means that the user agent interacts with the results analysis agent to determine what keywords/key phrases were present in a given result. The ODP-tree node annotation scheme presented in this chapter relies on keywords/key phrases to represent user interests. By com-bining user ratings (either explicitly or implicitly) with an analysis of the most frequently used keywords/key phrases in a document may be an indicator of how good (or bad) a certain key-word/key phrase is for describing a context in the ODP tree structure. Through the feedback process, the user agent can modify the relevance indicators stored at each leaf node in his/her individual ODP tree. This constitutes the refining of the agent’s belief about its user in the model presented in this section.
8.5 Conclusion
In this chapter, a discussion was rendered about the user agent in the COPEMSA model: the role it plays in the model and the techniques and strategies needed to help accomplish that role. The basis of any personalization effort by a computer system is the construction and maintenance of a user profile.
A tree-based strategy to user profiling was presented. The profiling strategy relies on the use of the open directory project (ODP) taxonomy for the definition of the contexts in which user queries can be classified. The approach also defines common and discriminating words that describe these contexts, and are stored as leaves in the ODP-tree.
The user’s primary interface with the user agent is through textual queries posted to the system.
The aspects surrounding this query interface and how context could be introduced into queries were also discussed.
In order for the user agent to continuously refine and collect data on its user(s), a method of stor-ing user profiles and a system of refinement must be considered. In this chapter, the concept of a user profile database was introduced for the storage of user profiles. The notions of an individual ODP-tree constituting the interests of the user and a user profile consisting of mappings between
user queries and nodes in the individual ODP-tree were also discussed.
Finally, the idea of feedback was introduced as the method of belief revision for the user agent of this chapter. Feedback may be collected either implicitly or explicitly. A combination of both strategies may however yield the best results in the context of a search system for the world wide web.
The personalization of a search system is one of the key features that could make the search en-gines of the future more effective in retrieving results that are more in line with user expectations and perceptions. The ideas and strategies presented in this chapter are merely a small step toward the ultimate goal of complete personalization of the search experience.
Chapter 9
The COPEMSA collaborative meta-search unit
In the previous chapter, a discussion was rendered about the capturing and personalization of user queries in the COPEMSA model. After a query has been submitted to the search system, the next logical step is the actual processing of the query and retrieval of results relevant to the submitted query.
In this chapter, the focus of discussion will shift to the collaborative meta-search unit of the COPEMSA model introduced in Chapter 7. The goals of the collaborative meta-search unit is twofold:
• The use of meta-search techniques for resource discovery1 from multiple search engines on the world wide web.
• Act as an interface for the rest of the system to a community of searchers, thereby leverag-ing on other’s search experiences.
The collaborative meta-search unit consists of two main components: the query agent and the community agent. The query agent is essentially, a scaled down version of a meta-search engine
1see the web mining process in chapter 6 for further details on the resource discovery step.
and its role in the model will be further elaborated on in this chapter. A brief introduction to some of the main challenges involved in the construction of a meta-search engine will also be presented.
The role of the community agent in the model will also be discussed, as well as some of the features that could be offered by collaboration between users in a search community.
9.1 Query agent
As has been mentioned earlier in this dissertation, the WWW contains a vast collection of in-formation and, in order to locate certain pieces of inin-formation, a large number of general and specialized search services has appeared on it. Additionally, no single search engine can index the entire web, giving rise to meta-search engines that attempt to provide a single unified access point to the multiple existing search engines on the web. Many systems that utilize meta-search techniques have been developed in recent years [81, 82, 83, 86, 87].
The query agent discussed in this section is in essence a scaled down version of a meta-search engine with the primary goal of locating relevant and useful results for a user’s query from multiple general and specialized search engines on the WWW. The place of the query agent in the suggested model as well as exactly how it can achieve the goal stated above is discussed in the following subsections.