Adaptive operator switching and solution space probability structure based genetic algorithm for information retrieval through pattern recognition

A novel adaptive operator set switching based genetic algorithm is proposed for pattern recognition based information retrieval problems. The problem the paper addresses specifically is the issue of large scale pattern recognition during the processes of information retrieval for e.g. in web search engines. The solution proposed is a novel genetic algorithm for the pattern recognition step, which utilizes the concepts of adaptive operator switching and the probability structure of the solution space. These two concepts have not been explored earlier within a single frame work by other researchers in the context of genetic algorithm for pattern recognition. The algorithm utilizes the probability structure of the solution space at every iteration to find better combinations of mutation and crossover operators. The operators are chosen from operator sets based on a mapping between the individual members of the sets and the probability distribution of the solutions. This mapping between the sets and distributions is changed as per the quality of the solutions at every iteration. This algorithm is highly scalable and is easily generalizable to strings, sets, sequences and graphs. This algorithm provides a significant improvement over some presently available optimization based algorithms for pattern recognition based information retrieval. The authors justify and validate the algorithm and its suitability through several theoretical arguments and inferences.