farhanahmadism / rule-based-nlu-for-voice-commands-detection

Rule-Based-NLU-for-Voice-Commands-Detection-

This is a rule based approach for Natural Language Understanding (NLU) for Voice Commands detection for Voice Assistants.

This is an approach for the solution of problem statement given by Samsung R&D Institute Bangalore during Hackfest 2018 at IIT(ISM) Dhanbad.

In this problem, various concept and grammar files were given and rules for each commands were given and we had to come up with an approach to detect the commands from the input sentence given by the user.For more detailed explanation of problem statement, refer to Samsung Problem statement1.doc and Samsung Problem statement1.doc files.

Our Approach - We basically used a binary string matching approach to detect the commands.We basically preprocessed all the grammar files of the various commands by making binary strings for each grammer of the commands given.We are making binary strings on the basis of which concept/placeholder is present there in that grammar of that specific command.Then, we made a binary string for the input command and matched with all the binary strings for all the commands and select that command which has the most match.Exact match is not required as we are following a probabilistic approach and trying to find that command which has the most match.

One of the problem was there with the open phrases which was somewhat solved bt was not working efficiently.Kindly refer to Samsung Problem statement2.doc for detailed explanation of the open phrases.