Property Recommendation

Objective

Given the current property that the user is browsing on the website, in this task, we aim to provide dynamic top k recommendations of ‘similar listings’ for the user.

Accomplishments

• We use two metrics to compare the quality of recommendations given by different recommenders, Alignment Score and the distance between Recommendation and Original Score.
• Data preprocessing to encode the oridnal and nonminal features then pass them through StandardScalar.
• Performed recommendation using KNN, KMeans, AGNES and DBScan.
• We improve the DBScan recommender by using KNN algorithm to find the top k recommendations if the item is identified as an outlier. With this improvement, the overall alignment score increased to 87.68% and the average distance between recommendations is reduced to 0.307.

Results

The overall best model we chose to implement the get top recommendations function is the KNN model. It has the highest alignment score of 90.64% and the shortest computation time.