Passage Retrieval en studie av index
The aim with this thesis came out of a strong interest for Passage Retrieval. Our intention has not been to evaluate an IR-system. Instead our goal has been to analyze the result of indexing documents and their passages. We have been studying the weights of the different terms in the different indices, in comparison with other parameters like frequency, normalized frequency and the inversed document frequency. Further more we have been looking at how the weights are spread using for instance the standard deviation. Our questions at issue are twofold. One: What differences are possible to find in an index after indexing whole documents and after indexing passages? Two: Is it possible to say that Passage Retrieval is more efficient exclusively by looking at the indices? We have investigated 98 documents and created five collections of passages. The indices have been large to break through. Our results have to a certain extent been predictable. We have found that the lowest weights always appear in the whole documents. As a rule of thumb we can see that as the passages reduce in size, the weights of the terms are growing. As to the other question, it is not possible to say whether Passage Retrieval could be preferred in relation to ordinary Information Retrieval, just studying indices.