In this project, we will build a model that automatically tags restaurants with multiple labels of the cuisine type using a dataset of user-submitted photos. Currently, restaurant labels are manually selected by Yelp users when they submit a review. Selecting the labels is optional, leaving some restaurants un- or only partially-categorized. With this data mining technique, the problem for un- or only partially-categorized restaurants will be solved. Also, users will have great experience where they don’t have to type the cuisine type.
Process:
-
Get the image dataset and business dataset from yelp data repository.
-
Load the photos and the business features.
-
Extract only those photos from the json data which are tagged with label - “Food”.
-
Use t-SNE/Truncated SVD/PCA and compare the results for feature extraction and dimensionality reduction on the input data.
-
Try dimensionality reduction before and after applying the variations to input set of images.
-
Create the different variations of each and every image and similarly keep track of the tag associated with original image with the newly generated images.
-
Pass the input data to the 5 layer of network i.e. 3 layers of Convolution feeding output of first to second and so on, flattening the output and then finally feeding it to fully connected layer.
-
The resulting Convolution Neural Network is used to train the model based on random weights and biases generated by the functions defined for it using the standard deviation according to resolution of images. 3000 iterations are run to finalize the weights and get an accurate model.
-
This model is then used to classify any image fed to it as input by the user.
-
Based on the image, a tag is identified, which in turn relates the image to a business id(Restaurant). This data is used to return a new set of tags and then using these tags we find the type of cuisine in the business.