This model describes how to develop a Doc2Vec (Distributed Memory Model) model in tensorflow. Lines 8-38 describes the model specification. If you wish to use pre-existing words such as Glove etc., change from word_embeddings = tf.Variable to word_embeddings = tf.placeholder. This is a modified version of word2vec from https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/udacity/5_word2vec.ipynb
The example does not currently have data to run. The user needs to provide the ids of the context words, and the id of the document as inputs. (eg. doc ids need to start at 0 and at len(docs)-1).
- Doc2Vec needs the document to be fairly large for the document vectors to be meaningful (capture a decent context of the words). This code was not meaningful with tweets. LDA worked better.
- You can capture more attributes like Author, Genre etc. with these vectors.