Possibility for intelligent feedback in online searching

Number of references containing the input term (set size) is the only feedback the searcher generally receives in most online information retrieval systems. There exist however, methods by which more intelligent feedback could be provided to the searcher. If the retrieved set of documents is not a very large one, computers could easily be made to select from this set the most informative features and feed them back to the terminal. These features can be indexing terms, classification codes, free‐text words, etc. A solution as to how these most informative features can be selected algorithmically in a database is presented here. The method is based on weighting terms properly: the weight depending on term frequency in the retrieved set and in the whole database. The selecting procedure can also be seen as a search for local semantic associations of search terms. What kind of feedback could be received by using this algorithm is simulated by sample searches from Inspec, Compendex, NTIS and Pascal databases. These examples show that useful synonyms for search strategy reformulation can automatically be found by this method.