Assume that the user has given a query "season win" and the system has returned as an answer documents d1, d6, and d8 from the document collection below. Next, the user has informed the system that document d1 is relevant but documents d6 and d8 are not. If the system uses relevance feedback, how could it modify the query? Which documents would the modified query return, if we use a matching principle that there should be at least one term such that the weight of the term is non-zero both in the query and the document. You can also try to order the answer documents, but you do not have to compute cosine similarity values etc. (unless you want to).
The weights do not have to be binary. They are set here to 1 and 0 (= " ") just to make the task simpler.
| contract | game | goal | hire | player | record | salary | sales | season | team | ticket | win | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| d1 | 1 | 1 | 1 | 1 | ||||||||
| d2 | 1 | 1 | ||||||||||
| d3 | 1 | 1 | 1 | 1 | ||||||||
| d4 | 1 | 1 | 1 | |||||||||
| d5 | 1 | 1 | 1 | 1 | ||||||||
| d6 | 1 | 1 | 1 | 1 | ||||||||
| d7 | 1 | 1 | 1 | |||||||||
| d8 | 1 | 1 |
In addition to relevance feedback, there are also other ways to expand user queries. Take a look of the references below and list some possibilities and how they can be combined with each other and with relevance feedback. You do not have to describe the whole contents of the material.
Simulate the KMP algorithm when
a) Simulate the BM algorithm when
b) Simulate the BM algorithm with a modification that only occurrence shift principle is used.