In this project, we created an end-to-end solution for
First, I built a clothing image classification model using a ResNet-based model.
The feature layer of this model can .... Then, using such features, the model can recommend similar stock index to the input stock using nearest neighbor search and also predict its price in short-term.
In the final phase, the project employs an autoencoder neural network to compress multidimensional stock market data into a lower-dimensional, encoded representation. This representation is used to calculate similarities between stocks, forming the basis of a content-based recommendation system. The system aligns with content-based filtering, recommending stocks based on their similarity in encoded data features.
To enhance accessibility, the recommendation system is deployed through a web application using Streamlit. Users can input specific stocks of interest and receive tailored recommendations from the sophisticated model.
Stock Market is a large-scale clothes database that is quite popular in the research community. It contains over 800,000 diverse fashion images ranging from well-posed shop images to unconstrained consumer photos. It is annotated with rich information of clothing items. It also contains over 300,000 cross-pose/cross-domain image pairs.
Market Index is
Financial News which we used raw_partner_headlines.csv
to directly-scraped raw news headlines. Columns go as follows: index, headline, URL, publisher, date, stock ticker.
- Phase I - EDA acquisition-and-EDA.ipynb: This code is used to pre-process the dataset.
- Phase II - Prediction Model & Sentiment Analysis model_stock_price_prediction.ipynb: This code is used to define all hyper-parameters regarding training.
- Phase III - Autoencoder DL Content-based Filtering & Web App Stock_Recommendation_System.ipynb: This code is used to pre-process the image further during training.