Conclusion

Challenges

Handling Large and Noisy Data:

Dealing with a massive and noisy dataset of over 300 GB, containing both product descriptions and reviews, posed a significant challenge for this project. We mitigated this issue by selecting a subset of the data and performing data cleaning to extract meaningful insights from it.

Implementing Personalized User Profiles:

Creating user profiles to offer personalized recommendations was a difficult task. However, we overcame this challenge by integrating an AI chatbot that interacts with users and collects information about their product preferences. By using this approach, we successfully built user-profiles and offered tailored recommendations.

Real-time Query Processing and Latency:

Processing user queries in real-time can cause latency issues, which was another challenge in this project. To address this problem, we pre-trained the model by generating product embeddings, review features, and image features ahead of time. Additionally, we divided the data into categories to enhance processing efficiency and reduce latency.

Future Steps

Scalable Collaborative Filtering using Entire Amazon Datasets:

We plan to implement scalable collaborative filtering techniques such as Matrix Factorization, Neural Collaborative Filtering, etc. to personalize recommendations for users by identifying similar users and recommending products based on their preferences.

Incorporating Text Data to Enhance Image-based Recommendation:

Currently, our recommender system relies on image data to identify similar products. However, we aim to improve the system's performance by incorporating text data of products such as product features, descriptions, etc., to identify visual features that can enhance the recommendations.

On-the-fly Aspect Based Sentiment Analysis for Personalized Recommendations:

We plan to overcome computational limitations and enhance the personalization of our recommender system by implementing on-the-fly Aspect-Based Sentiment Analysis. This will enable us to do ABSA on user-specific features and filter reviews based on them to provide more personalized and relevant recommendations.

Page updated

Google Sites

Report abuse