tappinganalytics/nlp

Use NLP to extract features from bodybuilding.com forum article titles and classify them into 8 muscle groups

NLP

Use NLP to extract features from EXRX.net exercise information and classify them into 8 muscle groups. Train the data with XGboost, Word2Vec, and Tensorflow.

Data

bodybuilding web crawler 67k data entries - 440MB exrx web crawler 308 data entries - 360kB

Muscle groups:

• Chest
• Biceps
• Triceps
• Glutes
• Back
• Leg
• Abs
• Shoulders

Pre-process data

• Remove numbers, punctuations, and https://
• TFIDFVectorizer

Model

• Benchmark : random forest
• XGboost : gradient boosting tree
• Tensorflow NN single layer
• Tensorflow RNN : recurrent neural network
• Word2Vec: gensim word2vec model

Required Libraries

• sklearn
• numpy
• pandas
• xgboost
• tensorflow
• json
• gensim

Result

Bodybuilding.com data are too generic for training purpose, we opted to use EXRX.net data. Tensorflow Neural Net gives 93% accuracy within its data set. We need to expand our model to predict potential user question and classify them into 8 categories. For this we use Word2Vec model for calculating the similarity between question sentence and muscle names. The accuracy so far is unclear.