https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/MSR_SCCD.pdf
Next word recommendation model for sentence completion using Deep Structured Semantic Models.
Research was based on:
https://www.researchgate.net/publication/262289160_Learning_deep_structured_semantic_models_for_web_search_using_clickthrough_data
trying to adopt same ideas applied to web search recommendations to sentence completion recommendations
Requirements:
tensorflow>=1.7
tf-hub
nltk
data: https://drive.google.com/drive/folders/0B5eGOMdyHn2mWDYtQzlQeGNKa2s
Problem: Current preprocessing method parses training data with distribution of sentance length far to the left (too short), while test data is distributed to the right