Information Retrieval 3 Retrieval model

The search model is divided into two types, one of which is the Boolean model ranked retrieval

A Boolean lookup:

1 binary Decision:is document relevant or not?

Documents are only relevant and unrelated, and are not ranked

2 presence of necessary and sufficient for match

All we need to do is record each document, and the words are OK.

3 operations that we query can have and and or both are set operations

Two ranked algorithm:

1 Frequency of document terms

The frequency of words appearing in the document, such as the appearance of Love in document 1 appeared 10 times, document 2 appeared love appeared 5 times,

So when the search term is love, document 1 should be ranked before document 2

2 Not all search terms necessarily present in document

This is mainly a Boolean to separate. For example, if you want to search for love girl then the file in the Boolean search must appear in both love and girl to be retrieved by us. And this is different ranked search, even if the document only Love is not girl will be searched, but the rankings will be slightly lower.

3 incarnations:

1) vector space model

2) Probabilistic model

3) Web search engines

Basis

Bag of words = like a set and also records a count for each element

Boolean model:

Search:

1 Specific search keywords

2 Boolean operators (and or not but xor) Note:xor = Exlusive or

Document:

Overall document collection forms Maximal Document Set

Advantages and Disadvantages

Documentation: Requires a high level of expertise to classify

Users: 1 users will not write a Boolean queries

2 users want to have relevance ranking----This is why the Boolean model is not suitable for web search

vector space Model:

Document:

1 document is expressed as bag of words

2 document is high-dimensional vector space

1 Each word is a direction

2 A variable of the frequency or frequency of a word is the value of a vector

Inquire:

Queries is also represented as a similar vector (for terms that exist in index)

Specific methods:

1 Select those documents that have the highest relevance to the query

2 Document-query similarity is the benchmark for ranking

3 The number of documents returned by the search becomes less important than the Boolean model, and the user looks down from top to bottom until satisfied

Score System:

1 distance

Problem freqency overweighted such as article A is not very important, but the appearance of a lot will also make the file ranking lower

2 cos

This is the way Vector-space model is used.

Through this article we have solved the problem of similarity between the Boolean model and how to measure vectors.

The next article will focus on how to manipulate the term specifically.

Information Retrieval 3 retrieval model