BIOMIMETIC SYSTEMS FOR INFORMATION RETRIEVAL

摘要

A powerful source of insight for computational intelligence is to examine how biological systems solve problems and then to construct analogous, biomimetic, mechanisms. Following this strategy, dolphin biosonar has been effectively modeled using neural networks. These networks perform brain-like computations, which are ideally suited to recognizing patterns. The same kinds of networks can also be applied to recognize the meaning patterns of words and documents. Biomimetic systems are self-organizing; they do not require the laborious construction of rigid, expensive, ontologies or prestructured rule bases. They discover the meaning of words in the same way that people do, by bootstrapping context. They allow true fuzzy semantic comparisons. The result is an ad hoc categorization system that adapts itself to the user's conceptual structure rather than forcing the user to adapt to the system. Humans constantly categorize, but their categories are unstable both from individual to individual and from time to time, depending on their needs and interests. Standard information retrieval approaches typically impose a single conceptual structure on their users. In contrast, biomimetic systems emulate the way their users' brains work and work in concert with them. Studies also find that people are poor at remembering the exact words that were used in a document, instead remembering the gist. As a result, document retrieval systems that depend on the presence of exact words fail to retrieve relevant documents. DolphinSearch technology, based on the biomimetic approach, learns the meanings of words from the documents it indexes and can recognize the relevance of documents based on their meaning.