An advanced search query based Recommendation System built using BERT based sentence_transformer model to recommendations for user input query. We implemented semantic search, question answering, summarization and document ranking and correcting spelling error in the search query.
- Recommend keywords that are more relevant to user’s search query by improving accuracy relevence and optimising the time taken.
- Improving the transactional impact by comparative visualization of the dataset scraped and the keywords/analytics provided. Then combining the relevant keys for a better output.
- Inserting extra features like input auto-correction, similar searches etc.
Stack: Python, Pandas, Numpy, Faiss, Seaborn, django, HTML, CSS
Model: Implementing Bert (Bidirectional Encoder Representations from Transformers) using Sentence Transformers
.jpg)
-
Crawling the given site and scraping the data through Scrapy.
Scraped Data : (2652,4)
- Pre-processing the scraped data along with keywords provided and then calculating frequency weightage i.e. finding most searched.
-
Implementing SBert model leading to :
a) Semantic Search (eg. insurance -> vehicle).
b) Paraphrase Mining (i.e. Text with identical/similar meaning).
c) Writing Search function -> Analyzing Query Vector ->Fetch into top k_ids.
