| 1 | 268 | Figure 7.5: Training Classifier
(In figure 7.5, the classifications are imagined to be part of a hierarchical classification system, as discussed in detail in Section 7.4.5 .) This data is used somehow (for now it's OK to think of it as magic:) to tune the set of parameters specifying a particular classifier. The second and dominant phase is then to use this classifier to automatically assign document to classes in an analogous manner to those in manually classified in the training set.
We seek the posterior probability of a particular class, given the evidence provided by a new document, |
| 2 | 266 | It is also important to remember that changes made to a single document in response to a single query can make no guarantees about improved performance with respect to other documents and other queries. For example, two documents might both be moved closer to a query (as proposed by Brauen/Roccio) while their relative rankings are not changed at all!
User drift and event tracking
One interesting feature of the training set generated by the routing task is the odd distribution of positive and negative examples it generates. Initially we can imagine that this filter is very inaccurate; |
| 3 | 182 | Inference beyond the Index
The Index that critical mapping between documents and descriptive keywords, has dominated our approach to FOA in all the preceeding chapters. But there is of course a larger context of available information: FOA can be accomplished by showing a user relations among keywords, by acquainting him or her with important authors, by pointing to important journals where relevant documents are often published, etc. Retrieval of all these information resources, especially when structured in meaningful interface, can tell a user much more than simply listing relevant documents.
This chapter is concerned with exploiting a variety of |
| 4 | 116 | Following a large number of such interactions documents which are wanted by the users will have been moved slowly into the active portion of the document space - that part in which large numbers of users' queries are concentrated, while items which are normally rejected will be located on the periphery of the space.
This provocative proposal, allowing a search engine to learn from its users}, is considered in much greater detail in Chapter 7.
Summary
We have been discussing RelFbk from the individual user's point of view. We've focused on how this information might be |
| 6 | 83 | is how best to quantify the discrimination power of the keyword.
Informative signals versus noise words
We begin with a weighting algorithm derived from information theory. Information theory has proven itself to be an extra-ordinarily useful model of many different situations in which some message must be communicated across a noisy channel and our goal is to devise an encoding for messages that is most robust in the face of this noise.
In our case, we must imagine that the messages describe the content of documents in our corpus. On this account, the amount of information |
| 7 | 163 | The J measure provides a criterion for retrieval function Match(). In experimental sitation the only preferences available are that Rel -Rel, but in natural retreval situations, users'richer RelFbk preference data can be used.
A particularly interesting use of this criterion is as part of error correction learning 7.3 . If we assume that the ranking function has certain free variables that we again have a training set of documents and that the criterion is differentiable with respect to a gradient search procedure can be used to adjust towards an optimal retrieval:
(Eq. 5.24)
For example, Bartell et al. consider |
| 8 | 155 | But many features of FOA suggest that index terms are highly dependent, highly correlated with one another. If that's the case, we can exploit that correlation by capturing only those axes of maximal variation and throwing away the rest.
Formal notions of similarity
Two features of the FOA problem can help us to focus on what is known as the Minkowski metric. First, the result of our calculations below will be a real-valued weight} associating a keyword with a document or query, and we can assume that this is a continuous quantity. Further, we can make the |
| 9 | 33 | users. The search engine users are searching for information about a topic they understand incompletely. Typical database users have a fairly specific question, like Query 3, in mind. It might even be that the database is missing some data; for example the special null value shows that the price of the third disk drive is not known. Even in this case, however, the database system knows that it doesn't know this information. FOA queries are rarely brought to such a sharp point; ambiguity is intrinsic to the users' expectations.
Because the queries are so general, an FOA retrieval must |
| 10 | 9 | used to count words. Even earlier, the related discipline of Library Science had developed many automated techniques for efficiently storing, cataloging and retrieving the physical materials so that browsing patrons could find them; many of these methods can be applied to the digital documents held within computers. IR has also borrowed heavily from the field of linguistics, especially computational linguistics.
The primary journals in the field and most important conferences Processing Management, the ACM's Transactions on Information Systems and the Journal of the American Society for Information Science (JASIS) are some of the central journals; meetings of the American |