Utilizing Hubel Wiesel models for semantic associations and topics extraction from unstructured text

There is a desire to extract and make better use of unstructured textual information available on the web. Semantic cognition opens new avenues in the utilization of this information. In this research, we extended the Hubel Wiesel model of hierarchical visual representation to extract semantic information from text. The unstructured text was preprocessed to a suitable input for Hubel Wiesel model. The threshold at each layer for neuronal growth was chosen as a ramp function of the level. Probabilistic approach was used for all post processing steps like prediction, word association, labeling, gist extraction etc. Equivalence with the Topics model was used to arrive at conditional probabilities in our model. We validated our model on three datasets and the model generated reasonable semantic associations. We evaluated the model based on top level clustering, label generation and word association.