A FAISS-based Search for Story Generation
Godwin C. George, Rajeev Rajan · 2022 IEEE 19th India Council International Conference (INDICON) · 2022
Stories have the power to change human perspectives and have applications in game development and film making. An intelligent system can generate appropriate stories for a set of keywords. We aim to build a system capable of getting stories by providing keywords as input. The stories must have a relation with the input keyword. We experimented with the ROCStory dataset. The preprocessed data are encoded using a sentence transformer, called msmarco-distilbert-base-prod-v3. We relied on a search approach based on Facebook AI Similarity Search (FAISS) to generate appropriate stories. The output story has been converted to audio via pyttsx3. The performance of the proposed model is compared with that of the sentence transformer paraphrase-MiniLM-L6-v2 approach. We made a subjective evaluation. The results show that the proposed approach outperforms the baseline method.