RelRec is a program utilizing the Yelp API to take a user's inputed geocoordinates and search details to return a sorted list of recommended nearby businesses. Program makes use of cosine similarity to identify businesses that need to be recommended and uses triadic closure for making friend recommendations based prior users' full look-up history of categories.
Social Networks
Physical Networks (Internet)
Recommendations
Advanced topics related to the class (privacy in social networks, recommendations, etc.)
Matthew Pickering - implemented all helper functions for getting and parsing API calls from Yelp API in YelpGetter.java. In addition, implemented Business.java and Category.java to handle responses from Yelp API. The most important implemented methods include getBusinesses(), getValidCategories(), getContentsFromRequest(), and parseJsonStringToObject(). Finally, implemented static variables at top of YelpGetter to permit easy testing and building of requests.
Tinatin Kokoshvili - implemented getReviews() function that gets reviews of appropriate businesses from Yelp API.
implemented Recommendations class, which handles GUI and user interaction, does cosine similarity
of user's description and reviews of business, creates vectors of user and businesses with appropriate
entries, computed angle between them, sorts the businesses, and displays top (up to 5) businesses to the user.
Nancy Zhu - implemented users class that gets the category and name that the user types into the GUI text fields and then adds them to a history log for the user stored under txt files. The users class provides friend recommendations based on the triadic closure property where the most frequent category search history is the common characteristic between the user and recommended friend.
Check user manual pdf.
After downloading the most current version of the repository, enter the 'src' folder and run Recommendations.java. A panel should pop-up. The input is
-
Name Uesr's name; name field should not be left empty
-
Category Valid input: any Yelp category Link to yelp categories: https://www.yelp.com/developers/documentation/v3/all_category_list Category field should not be left empty
-
Price range Valid inputs: 0, 1, 2, 3, 4 Number corresponds to number of dollar(or other currency) symbols for businesses 0 being least expensive, and 4 being most expensive If field is left empty, default value will be used, which is 0.
-
Description Valid Input: any number of words that describe the business or category For example, "tasty, cosy, friendly, etc..." This field should not be left empty.
-
latitude Valid input: any latitude within countries where Yelp has access to businesses For example, random location is Rome: 41.8933203 This field should not be left empty.
-
longitude Valid input: any longitude within countries where Yelp has access to businesses For example, random location is Rome: 12.4829321 This field should not be left empty.
-
Distance Valid input: any number Note that the distance is in meters. If the user leaves it empty, the default value will be used, which is 0.
-
Radius Valid input: any number Note that the radius is in meters. If field is left empty, the default value will be used, which is 30000
-
Rating Valid input: 0, 1, 2, 3, 4, 5 If left empty, default value will be used, which is 5.
After giving input and pressing "Get Recommendations" button, user will be able to see the pop-up that includes up to five recommended businesses. Below businesses, there are friend recommendations based on the prior category searches of other users. Note that the program might take a while to run. Also, sometimes we get an error from server when program makes API calls. Please, if this happens just re-press the "Get Recommendations" button. It should fix it.
For the time being, the src folder includes the following classes:
Good ol' URLGetter from the previous homework coded by our good friend Swap. This is used essentially just to build the HTTPURLConnection.
Defines a Business class to handle all necessary fields, constructors, and methods that define a business we may recommend. Includes fields such as pricing, rating, rating count, location, categories, etc. As of now, this class encodes the backbone data structure for how we handle businesses.
Defines a Category class to handle all necessary fields, constructors, and methods that define a category to input or output to/from the Yelp API. Each carries the categories title (interpretable by humans), alias (essentially variable name interpretable by Yelp API), and set of parent categories (ex: 'Food' is a parent category of 'Pizza'). Created to allow easily handling on our end of the categories a business holds.
Designs a YelpGetter class to handle creation and execution of API calls to the Yelp API. Has an overloaded constructor allowing for the creation of an API call with or without a specified category to search within. However, every created YelpGetter.java must be called with an inputted latitude, longitude, and search radius. Contains several helper functions managing what constitutes a legal category, how to parse out a received JSON file, and how to get the JSON file associated with a URL request.
At a high level, YelpGetter.java functions with just one field: the URL to follow to interact with the Yelp API. Since every API call to the Yelp API must include a latitude and longitude, those are necessary elements to construct the YelpGetter. The radius is an added necessary input to create the URL because I believe the radius of the businesses to get is a necessary feature. The category is an optional element of the URL because the Yelp API allows the making of calls that are and aren't specified by category. Once a YelpGetter is created and the URL field is created, one can call the getBusinesses() method to retrieve all businesses returned by the Yelp API for that particular request.
We also make calls to API to get reviews of all the businesses appropriate to category, using getReviews() method.
GUI Panel is created with all the necessary TextFields. HintTextField class is created within this java file. It shows a hint if the TextField is not clicked. If the user clicks it it to type something, hint will disappear. Main panel includes a "Get Recommendations" button that shows a MessageDialog and displays recommendations for businesses and friends.
Cosine Similarity for Description and reviews We make use of a slightly modified VectorSpaceModel.java file from the class. We find cosine similarity between user's description of a place and reviews of a business. This number describing similarity to user's description is used as one entry in the final business vector in the following step.
Cosine Similarity between user and business vectors
To find the businesses that match the user input the most, we create user and business vectors that
include entries: price, distance, rating, and reviews cosine similarity to user description.
We compute angles between user and all other business vectors and recommend businesses
that make the smallest angles with user vector.
Class creates angleBetween() method that computes and angle between
two vectors.
Most of the above logic is included in getRecommendations() method.
Defines a users class that delivers friend recommendations to the user based on triadic closure of the most frequent category searches. The class creates methods that reads into txt files that represent every user that has logged into the application. The txt files are read and a hash map that creates a frequency chart of the category and the frequency of those user's category searches is created. The frequencies are then compared against the user's most frequent category and then using triadic closure, a friend recommendation based on the most frequent category searches is generated and then returned.
We take advantage of two external libraries:
- Reverse Country Code by bencampion Used under the Apache-2.0 License
- Gson by Google Inc. Used under the Apache-2.0 License
VectorSpace Model from class.