System and method for infusing knowledge graphs and language models for natural language sentence pair applications
The present disclosure relates to a system for infusing knowledge graphs and language models (LM) for natural language sentence pair tasks, the system include a processor operatively coupled to an inference engine, the inference engine configured to receive a sentence indicative of a premise and a sentence indicative of a hypothesis, extract LM embeddings for the corresponding sentence, generate a common-sense knowledge graph for corresponding sentence and nodes are derived from the common-sense knowledge graph, assign node importance scores for each of the derived nodes, compute node parameters for the derived nodes, apply an aggregation function to generate pooled values, concatenate the LM embeddings, graph embeddings and pooled values to generate concatenated data and classify the concatenation data to indicate a relationship between the natural language inference pair.
1. A system for infusing knowledge graphs and trained language models (LM) for natural language sentence pair tasks, said system comprising:
a processor operatively coupled to an inference engine, the processor operatively coupled to a memory, the memory storing instructions executable by the processor to:
receive a sentence indicative of a premise and a sentence indicative of a hypothesis, wherein the sentence indicative of the premise and the sentence indicative of the hypothesis form a natural language inference classification pair;
process the sentence indicative of the premise and the sentence indicative of the hypothesis to extract trained LM embeddings of the sentence indicative of the premise and the sentence indicative of the hypothesis;
generate a common-sense knowledge graph for corresponding sentence to extract premise sub-graphs and hypothesis sub-graphs using the knowledge graph, wherein nodes are derived from the common-sense knowledge graph;
assign node importance scores for each of the derived nodes using trained LM embeddings;
compute, by a graph attention network (GAT), node parameters for each of the derived nodes, apply an aggregation function to generate pooled values;
concatenate the trained LM embeddings, graph embeddings and pooled values of graph to generate concatenated data; and
classify the concatenation data, wherein the output of classifying the concatenation data indicates a relationship between the natural language inference pair.
2. The system as claimed in claim 1 , wherein the node importance scores are computed using pre-trained language model taking the premise sub-graphs and hypothesis sub-graphs as two separate contexts while the nodes remain the same for both the subgraphs.
3. The system as claimed in claim 1 , wherein the node importance scores is assigned for each of the derived nodes, wherein the derived nodes comprise premise nodes, hypothesis nodes and common nodes.
4. The system as claimed in claim 3 , wherein the common nodes formed based on the relationship with the premise nodes and the hypothesis nodes.
5. The system as claimed in claim 1 , wherein the node parameters comprise node type, relation between nodes, and node importance scores.
6. The system as claimed in claim 1 , wherein the common-sense knowledge graph is employed for natural language processing (NLP) tasks comprising natural language inference (NLI) and semantic textual similarity (STS).
7. The system as claimed in claim 1 , wherein the system determines similarities and differences between the sentences for natural language inference (NLI) requirements.
8. The system as claimed in claim 1 , wherein the system computes cosine similarity to calculate semantic similarity between the sentences satisfying semantic textual similarity (STS) requirements.
9. A method for infusing knowledge graphs and trained language models for natural language sentence pair tasks, the method comprising:
receiving, at a computing device, a sentence indicative of a premise and a sentence indicative of a hypothesis, wherein the sentence indicative of the premise and the sentence indicative of the hypothesis form a natural language inference classification pair;
extracting, at the computing device, trained LM embeddings of the sentence indicative of the premise and the sentence indicative of the hypothesis;
generating, at the computing device, a common-sense knowledge graph to extract premise sub-graphs and hypothesis sub-graphs using the common-sense knowledge graph, wherein nodes are derived from the common-sense knowledge graph;
assigning, at the computing device, node importance scores for each of the derived nodes using trained LM embeddings;
computing, at the computing device, node parameters for each of the derived nodes and applying an aggregation function to generate pooled values;
concatenating, at the computing device, the trained LM embeddings, graph embeddings and pooled values of graph to generate concatenated data; and
classifying, at the computing device, the concatenation data, wherein the output of classifying the concatenation data indicates a relationship between the natural language inference pair.