Temporal Case-Based Reasoning Retrieval Through Transformers Using Context-Based Vocabularies

In the field of problem-solving, case-based reasoning (CBR) emerges as a technique that leverages past experiences to address new challenges based on the assumption that there is a pattern that can be exploited. The temporal variant of case-based reasoning, known as TCBR (Temporal Case-based Reasoning), involves the application of CBR to time-varying inputs, finding typical applications in medical and industrial contexts. Over the years, various approaches have been explored to deal with the temporal nature of these problems. The goal of this research is to explore additional techniques associated with the case recovery phase in CBR, given the existence of multiple methods not definitively established in the literature. This paper focuses on using transformers, a modern neural network model designed for time-varying problems, as a means of case retrieval. Initial tests are presented where successful case retrieval based on discretized sequences of inputs is achieved. In addition, further steps are proposed to deepen the potential use of this technique. Integrating the concept of continuous improvement, the notion of “Context-Based Vocabularies” is introduced, which seeks to enrich case retrieval capabilities in changing temporal situations. These dynamic vocabularies, adapted to the temporal context, aim to strengthen the CBR system's ability to identify patterns and perform more accurate retrievals in environments where temporal variability is a crucial factor. This approach represents a step forward in the evolution of CBR strategies, highlighting the importance of adaptability and continuous improvement in effective problem-solving in dynamic contexts.