This is the implementation of the runtime phase of the pipeline for semantics-driven keyword search over knowledge graphs proposed by L.Feddoul. This code version uses the cosine similarity between the embeddings of the node labels as a semantic relatedness metric. The embeddings are generated using the Universal Sentence Encoder lite. This pipeline needs a weighted summary graph as input generated by the off-line phase. To run the code, we provide an example summary graph of Mondial under `query-generation-code/summary-graph/mondial` that is generated by running the off-line phase of the pipeline.