Learned Recommendations on Yelp

Recommendations are suggestions that may direct the target toward something that they may like. In Yelp, data analysis is required in order to make satisfactory recommendations to users. Recommendations should thereby be supervised when they are being made so that each can be specific to their user. There are many proposed ways to receive recommendations on Yelp. The challenge here is that there is an abundance of information and it is hard to determine the importance of each relationship a user has (Yu, et al.). Therefore, there are many ways to give suggestions for which places to eat or go to.

One possible system for recommendations involves using a social network. By looking at what users like most frequently, we can create a list of recommendations just from that. This can be particularly helpful for new users as they will not have much history that Yelp can train recommendations from (Qian, et al.). Furthermore, if there is data and social networking data, a personalized system using location can also be implemented to generate a more accurate reading of a user’s inclinations (Savage, et al.). Another method involves a Spatial Topic which compares location and interests and see if there are functions of interest at those locations (Hu and Ester). It’s interesting to note that those who live near each other are likely to have similar movement behaviors. For example, students at the New York University may go to the same restaurants and grocery stores due to equal relative proximity.

Figure 1: Finding number of contextual variables from a number of words (Bauman, et al.)

An interesting method for a recommendation system uses the text that a user writes for their reviews and distinguishes any contextual variables in the text (Bauman, et al.). Furthermore, it is important to discern which of the reviews in the data set are specific and which are generic. By mining variables from the specific reviews, we have more reliable results to work with. Then, using the word-based method will give us contextual variables that basically give the context for a review. This works by analyzing which words have high frequency in a specific review. This will also give insight into the user and allow for better recommendations to be made.

Works Cited

Bauman, Konstantin, and Alexander Tuzhilin. “Discovering Contextual Information from User Reviews for Recommendation Purposes.” CBRecSys@ RecSys. 2014.

Hu, Bo, and Martin Ester. “Spatial topic modeling in online social media for location recommendation.” Proceedings of the 7th ACM conference on Recommender systems. ACM, 2013.

Qian, Xueming, et al. “Personalized recommendation combining user interest and social circle.” IEEE transactions on knowledge and data engineering26.7 (2014): 1763-1777.

Savage, Norma Saiph, et al. “I’m feeling loco: A location based context aware recommendation system.” Advances in Location-Based Services. Springer Berlin Heidelberg, 2012. 37-54.

Yu, Xiao, et al. “Recommendation in heterogeneous information networks with implicit user feedback.” Proceedings of the 7th ACM conference on Recommender systems. ACM, 2013.