Methods and systems for establishing semantic equivalence in access sequences using sentence embeddings

Abstract

Systems and methods are provided for utilizing natural language process (NLP), namely semantic learning approaches in network security. Techniques include analyzing network transaction records to form a corpus related to a semantics of network activity. The corpus includes formulated network sentences, representing sequences of network entities that are accessed in the network. A corpus of network sentences can include sequences of servers accessed by each user. A network sentence embeddings model can be trained on the corpus. The network sentence embeddings model includes an embedding space of text that captures the semantic meanings of the network sentences. In sentence embeddings, network sentences with equivalent semantic meanings are co-located in the embeddings space. Further, proximity measures in the embedding space can be used to identify whether network sentences (e.g., access sequences), are semantically equivalent. Using network sentence embeddings model, equivalent semantics of access can be established to efficiently detect anomalies.