Prediciting restaurant and popularity based on Yelp Dataset project report
- 1.    Predicting Restaurants Rating and Popularity based on Yelp Dataset   MACHINE LEARNING PROJECT REPORT      Submitted by ALIN BABU (67) NANDU O (66) LIJU THOMAS (36)    Â
- 2.   Introduction  Restaurants rating on Yelp becomes an important indicator of their future. In this             project, we focus on predicting ratings and popularity change of restaurants. With data             from Yelp, we use several machine learning methods including logistic regression and            Naive Bayes, to make relevant predictions. While logistic regression seems to perform            better than the others, predictions from all the methods are far from perfect. This implies               the potential improvement of more data and more suited methodology.  Project Objectives  â To predict ratings of restaurants on Yelp and popularity change based on            restaurant features. â Project can shed light on what customers value the most about a restaurant.  Dataset â The data comes from Yelp Dataset Challenge . â It includes review data, including text, time and rating. â From the raw dataset, we select 20000 samples for testing. â Due to different cultures across cities, we only focus on restaurants in a particular              city and surrounding areas in this project. 1Â
- 3.   Algorithm and Methods   In this project we use mainly three machine learning algorithms to predict the restaurant rating.The algorithm used here are supervisory learning .The algorithm used in the project are: â Logistic Regression â Multinomial Naive Bayes â Naive Bayes Logistic Regression Logistic regression is a classification algorithm used to assign observations to a            discrete set of classes. Unlike linear regression which outputs continuous number values,            logistic regression transforms its output using the logistic sigmoid function to return a             probability value which can then be mapped to two or more discrete classes. Naive Bayes  A Naive Bayes classifier is a probabilistic machine learning model thatâs used for             classification task. The crux of the classifier is based on the Bayes theorem. Bayes Theorem: 2Â
- 4.  Using Bayes theorem, we can find the probability of A happening, given that B has               occurred. Here, B is the evidence and A is the hypothesis. The assumption made here is                that the predictors/features are independent. That is presence of one particular feature            does not affect the other. Hence it is called naive. Multinomial Naive Bayes Multinomial Naive Bayes is a specialized version of Naive Bayes that is designed             more for text documents. Whereas simple naive Bayes would model a document as the              presence and absence of particular words, multinomial naive bayes explicitly models the            word counts and adjusts the underlying calculations to deal with in.    Data Pre-Processing In this project we use mainly Yelp Dataset.Dataset consists of user review and             rating. We mainly use restaurant name, date, comfortability, star rating, comments,           review id out of these data we will two features which are essential for our prediction. Selecting two valid features manually. â Star rating â Comments After selecting the valid features we find the missing values of an attribute and then root word extracting from comment rating using the methods below: â Removing punctuations 3Â
- 5.  â Removing stop words â Stemming - The process of producing morphological variants of a root/base word.   Performance Evaluation  Logistic Regression   Naive Bayes   4Â
- 6.   Multinomial Naive Bayes    Conclusion   After testing with 20000 samples we can see that logistic regression performs            better than the other methods. One possible explanation is that the assumptions for other              models are problematic, and logistic regression is more robust to problematic model            assumptions.This implies the potential improvement of more data and more suited           methodology.However, the prediction needs further improvement. We compare our best          predictor-logistic regression with a random-number predictor, and a constant-number         predictor. As we can see, the logistic predictor is only slightly better than the              constant-number predictor.  5Â