Movie Recommendation from Conversational Data: ConvExtr

Re-implementation and extension of the ACL paper You Sound Like Someone Who Watches Drama Movies: Towards Predicting Movie Preferences from Conversational Interactions

---

Work Done

• Built a movie recommendation system leveraging user conversational data (transcripts), critics data and domain adaptation techniques
• Obtained a 3% improvement on existing results of the paper and performed hyperparameter tuning on all the three CF approaches: KNN, SVD and SVDpp.
• Experimented with neural matrix factorization approaches as an extension of the paper and obtained comparable results of RMSE=1.232 and MAE=0.9569.

---

Tech Stack

• Python 3
• surprise
• Microsoft Recommenders Framework
• seaborn

---

Report and Presentation

You can access the report here and presentation here.

Structure and Acknowledgements

The file "NLP_Project_Step1.ipynb" contains code for the re-implementation and hyperparameter tuning to obtain results published in the paper. The file "NLP_Project_Step2.ipynb" also contains code where the NMF extension is applied.

Made as a team with @janish-parikh and Jash Gaglani.