The DENDRAL Project was one of the first large-scale programs to embody the strategy of using detailed, task-specific knowledge about a problem domain as a source of heuristics, and to seek generality through automating the acquisition of such knowledge. This paper summarizes the major conceptual contributions and accomplishments of that project. It is an attempt to distill from this research the lessons that are of importance to artificial intelligence research and to provide a record of the final status of two decades of work.

Don R. Swanson has undertaken a program of research to use the published medical literature as a source of discoveries. We have attempted to replicate his discovery of a connection between Raynaud's disease and dietary fish oil, as well as develop computer-based searching methods that could usefully support literature-based discoveries. We have been successful in replicating Swanson's discovery and have developed a method of discovery support based on the complete text of MEDLINE records. From these, we compute statistics based both on the frequency of tokens within a literature and on the number of records containing various tokens. We discuss the use of these statistics, suggesting that token and record frequencies are good indicators of literatures profitably related to some source literature, and that relative record frequencies are useful in isolating literatures with the potential of containing a discovery.

We report experiments that use lexical statistics, such as word frequency counts, to discover hidden connections in the medical literature. Hidden connections are those that are unlikely to be found by examination of bibliographic citations or the use of standard indexing methods and yet establish a relationship between topics that might profitably be explored by scientific research. Our experiments were conducted with the MEDLINE medical literature database and follow and extend the work of Swanson.

Don R. Swanson has undertaken a program of research to use the published medical literature as a source of discoveries. We have attempted to replicate his discovery of a connection between Raynaud's disease and dietary fish oil, as well as develop computer-based searching methods that could usefully support literature-based discoveries. We have been successful in replicating Swanson's discovery and have developed a method of discovery support based on the complete text of MEDLINE records. From these, we compute statistics based both on the frequency of tokens within a literature and on the number of records containing various tokens. We discuss the use of these statistics, suggesting that token and record frequencies are good indicators of literatures profitably related to some source literature, and that relative record frequencies are useful in isolating literatures with the potential of containing a discovery.

Previous research has shown that researchers can generate medical hypotheses by using computers to analyze several, seemingly unrelated, medical literatures. In this work we suggest broader application for the ideas of literature-based discovery. Specifically, we suggest that literature-based discovery can be fruitful in areas other than medicine; that in addition to finding "cures" for "problems," literature-based discovery offers the possibility of finding new problems for existing technologies; that the analysis of a single literature may be sufficient for literature-based discovery; and that literature-based discovery can support individuals seeking to draw together ideas from various areas of inquiry, even if such connections have been previously made by others.We describe literature-based discovery experiments conducted on the World Wide Web that support these ideas.

No abstract

2 experiments were conducted in which Ss gave 1 of 3 responses to each stimulus in a random sequence prepared from 32 distinct stimuli which assumed 1 of 2 levels for each of S dimensions. The sequence was constructed so that 2 of the 32 stimuli occurred with probability $ each and the remaining 30 stimuli occurred with probabilities summing to J. Ss were instructed to respond by depressing a -key to one of the high frequency stimuli, and a + key to the other, and a 0 key to any of the remaining 30. Results support the hypothesis that frequently occurring stimuli may be identified as total patterns, perhaps by some sort of template matching which compares all dimensions simultaneously. However, the results also suggest that the template matches are made serially, and infrequent (or unfamiliar) stimuli are identified by a serial examination of stimulus dimensions.

This article compares the traditional, fixed problem representation style of a genetic algorithm (GA) with a new floating representation in which the building blocks of a problem are not fixed at specific locations on the individuals of the population. In addition, the effects of noncoding segments on both of these representations is studied. Noncoding segments are a computational model of noncoding deoxyribonucleic acid, and floating building blocks mimic the location independence of genes. The fact that these structures are prevalent in natural genetic systems suggests that they may provide some advantages to the evolutionary process. Our results show that there is a significant difference in how GAs solve a problem in the fixed and floating representations. Genetic algorithms are able to maintain a more diverse population with the floating representation. The combination of noncoding segments and floating building blocks appears to encourage a GA to take advantage of its parallel search and recombination abilities.