DualNAVI manual

Basic

Details

Customization

Technical Points

Concept

Associative search is a type of information retrieval based on the similarity between documents.

Associative search is useful when the user's information need is not clearly expressed by one or several keywords, but the user has some interesting texts (see Clipboard Association.)

Associative search is also useful as a feedback tool. If you find interesting items in the search results, then associative search with these items as search keys may bring you more related items which were not previously retrieved. (see Interaction in Basic.)

Associative search in DualNAVI consists of following steps:

• Extraction of characterising words from the selected documents.
  • The default number of characterising words is set at 200.
    • This value is set as the Number of Internal Words of PE in the Advanced section of Option window.
  • For each word (w), which appears at least once in the selected documents, its score is calculated by score(w) = tf(w) / TF (w), where tf(w) and TF(w) are the term frequencies of w in the selected documents, and in the whole database respectively.
    Then the above number of words in the order of this score are selected.
• These extracted words are used as a query, and the relevance of each document (d) in the target database with this query (q) is calculated by the following formula: (where DF(w) is the number of documents containing w, and N is the totla number of documents in the database.)
  • 
• The results are sorted by this similarity and the top scored documents are returned as the search results.
  • The number of search results is set as the Number of Articles in the Basic section of Option window. (See customization.)

Topic word graphs summarize the search results and suggest proper words for further search.

• The words composing a topic word graph are those characterizing the retrieved documents.
  • The score for selecting topic words is given by df(w) / DF(w), where df(w) is the number of retrieved documents containing w and DF(w) is total number of documents (in the database) containing w.
  • In order to make balanced selection of common words and specific words, all the candidate words (those appearing in at least one retrieved document) are classified into several frequency classes, and proper number of topic words are selected from each frequency class.
• A link (an edge) between two words means that the two words are strongly related, that is, they co-appear in many documents in the retrieval results.
  • Each topic word X is linked to another topic word Y which maximizes the co-occurence strength df(X & Y) / df(Y) with X among those having higher document frequency than X. Here df(X & Y) means the number of retrieved documents which have both X and Y.
  • The length of a link does not mean anything. (Althogh it may be natural to expect a shorter link means a stronger relation.)
• As for the two dimensional arrangement of topic words, the vertical coordinate means the (document) frequency of each word in the retrieved documents, whereas the horizontal coordinate has no meanings.
  High frequency words (common words) are placed in the upper part, and low frequency words (specific words) are place in the lower part. Therefore, this graph can be considered as a hierarchy of topics in the search results.