Sentiment analysis : Machine-Learning approach

Following up on my earlier post, as the frequency-based models were not very accurate and a good rule-based model was very hard to elaborate, we implemented what we known to be state-of-the-art methods for sentiment analysis on short sentences and make a list of the pros and cons of these methods. We train all of them on a 10.000 sentences dataset. These sentences are classified as positive, neutral, and negative by human experts. We the benchmark the models on a hold out sample of 500 sentences.

Word representations in a vector space

Feature extraction

To build a deep-learning model for sentiment analysis, we first have to represent our sentences in a vector space. We studied frequency-based methods in a previous post. They represent a sentence either by a bag-of-words, which is a list of the words that appear in the sentence with their frequencies, or by a term frequency – inverse document frequency (tf-idf) vector where the word frequencies in our sentences are weighted with their frequencies in the entire corpus.

These methods are very useful for long texts. For example, we can describe very precisely a newspaper article or a book by its most frequent words. However, for very short sentences, it’s not accurate at all. First, because 10 words are not enough to aggregate. But also because the structure of the sentence is very important to analyze sentiment and tf-idf models hardly capture negations, amplifications, and concessions. For instance, “Very good food, but bad for service…” would have the same representation as “Bad for food, but very good service!”.

Word vectors

We represent our sentences with vectors that take into account both the words that appear and the semantic structure. A first way to do this is to represent every word with an n-feature vector, and to represent our sentence with a n*length matrix. We can for instance build a vector of the same size as the vocabulary (10.000 for instance), and to represent the i-th word with a 1 in the i-th position and 0 elsewhere.

Tomas Mikolov developed another way to represent words in a vector space, with features that capture the semantic compositionality. He trains the following neural network on a very large corpus:

Neural network trained to get Word2Vec's word vectors

He trains this model and represents the word “ants” by the output vector of the hidden layer. The features of these word vectors we obtain capture most of the semantic information, because it captures enough information to evaluate the statistical repartition of the word that follows “ants” in a sentence.

What we do is similar. We represent every word by an index vector. And we integrate in our deep learning model a hidden layer of linear neurons that transforms these big vectors into much smaller ones. We take these smaller vectors as an input of a convolutional neural network. We train the model as a whole, so that the word vectors we use are trained to fit the sentiment information of the words, i.e. so that the features we get capture enough information on the words to predict the sentiment of the sentence.

Sentence representations


We want to build a representation of a sentence that takes into account not only the words that appear, but also the sentence’s semantic structure. The easiest way to do this is to superpose these word vectors and build a matrix that represents the sentence. There is another way to do it, that was also developed by Tomas Mikolov and is usually called Doc2Vec.

He modifies the neural network we used for Word2Vec, and takes as an input both the word vectors that come before, and a vector that depends on the sentence they are in. We will take the features of this word vector as parameters of our model and optimize them using a gradient descent. Doing that, we will have for every sentence a set of features that represent the structure of the sentence. These features capture most of the useful information on how the words follow each other.

Neural Network trained to get Doc2Vec's document vectors

Pros and cons for sentiment analysis

These document vectors are very useful for us, because the sentiment of a sentence can be deduced very precisely from these semantic features . As a matter of fact, users writing reviews with positive or negative sentiments will have completely different ways of composing the words. Feeding a logistic regression with these vectors and training the regression to predict sentiment is known to be one of the best methods for sentiment analysis, both for fine-grained (Very negative / Negative / Neutral / Positive / Very positive) and for more general Negative / Positive classification.

We implemented and benchmarked such a method but we chose not to productionalize it. As a matter of fact, building the document vector of a sentence is not an easy operation. For every sentence, we have to run a gradient descent in order to find the right coefficients for this vector. Compared to our other methods for sentiment analysis, where the preprocessing is a very short algorithm (a matter of milliseconds) and the evaluation is almost instantaneous, Doc2Vec classification requires a significant hardware investment and/or takes much longer to process. Before taking that leap, we decided to explore representing our sentences by a matrix of word vectors and to classify sentiments using a deep learning model.

Convolutional neural networks

Convolutional neural networks

The next method we explored for sentiment classification uses a multi-layer neural network with a convolutional layer, multiple dense layers of neurons with a sigmoid activation function, and additional layers designed to prevent overfitting. We explained how convolutional layers work in a previous article. It is a technique that was designed for computer vision, and that improves the accuracy of most image classification and object detection models.

The idea is to apply convolutions to the image with a set of filters, and to take the new images it produces as inputs of the next layer. Depending on the filter we apply, the output image will either capture the edges, or smooth it, or sharpen the key patterns. Training the filter’s coefficients will help our model build extremely relevant features to feed the next layers. These features work like local patches that learn compositionality. During the training, it will automatically learn the best patches depending on the classification problem we want to solve. The features it learns will be location-invariant. It will convolve exactly the same way an object that is at the bottom of the frame and an object that is at the top of the frame. This is key not only for object detection, but for sentiment analysis as well.

Convolution used for edge detection

Convolution used for edge detection

Applications in Natural Language Processing

As these models became more and more popular in computer vision, a lot of people tried to apply them in other fields. They had significantly good results in speech recognition and in natural language processing. In speech recognition, the trick is to build the frequency intensity distribution of the signal for every timestamp and to convolve these images.

For NLP tasks like sentiment analysis, we do something very similar. We build word vectors and convolve the image built by juxtaposing these vectors in order to build relevant features.

Intuitively, the filters will enable us to highlight the intensely positive or intensely negative words. They will enable us to understand the relation between negations and what follows, and things like that. It will capture relevant information about how the words follow each other. It will also learn particular words or n-grams that bear sentiment information. We then feed a fully connected deep neural network with the outputs of these convolutions. It selects the best of these features in order to classify the sentiment of the sentence. The results on our datasets are pretty good.

Convolutional neural networks for Natural Language Processing


We also studied, implemented and benchmarked the Long Short-Term Memory Recurrent Neural Network model. It has a very interesting architecture to process natural language. It works exactly as we do. It reads the sentence from the first word to the last one. And it tries to figure out the sentiment after each step. For example, for the sentence “The food sucks, the wine was worse.”. It will read “The”, then “food”, then “sucks”, “the” and “wine”. It will keep in mind both a vector that represents what came before (memory) and a partial output. For instance, it will already think that the sentence is negative halfway through. Then it will continue to update as it processes more data.

Recurrent neural networks - The human way to do sentiment analysis

This is the general idea, but the implementation of these networks is much more complex because it is easy to keep recent information in mind, but very difficult to have a model that captures most of the useful long-term dependencies while avoiding the problems linked to vanishing gradient.

This RNN structure looks very accurate for sentiment analysis tasks. It performs well for speech recognition and for translation. However, it slows down the evaluation process considerably and doesn’t improve accuracy that much in our application so should be implemented with care.

Sentiment trees – RNTN model

Richard Socher et al. describe in the paper Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank another cool method for sentiment analysis. He says that every word has a sentiment meaning. The structure of the sentence should enable us to compose these sentiments in order to get the overall sentiment of the sentence.

Stanford Sentiment Treebank example

They implement a model called the RNTN. It represents the words by vectors and takes a class of tensor-multiplication-based mathematical functions to describe compositionality. Stanford has a very large corpus of movie reviews turned into trees by their NLP libraries. Every node is classified from very negative to very positive by a human annotator. They trained the RNTN model on this corpus, and got very good results. Unfortunately, they train it on IMDB movie reviews data. But it doesn’t perform quite as well on our reviews.

The big advantage of this model is that it is very interpretable. We can understand very precisely how it works. We can visualize which words it detects to be positive or negative, and how it understands the compositions. However, we need to build an extremely large training set (around 10.000 sentences with fine-grain annotations on every node) for every specific application. As we continue to gather more and more detailed training data, this is just one of the types of models we are exploring to continue improving the sentiment models we have in production!

Sentiment analysis : Frequency-based models

We give our tenants insights about their online reputation based on their online reviews and ratings. In doing so, one thing we try to do is pull apart the text of reviews to understand what the reviews are dealing with, and tell our clients what their customers are talking about and how happy those customers are with key aspects of our clients’ business.

So for example, we might identify 100 reviews for our client mentioning price, and leveraging the star rating of those reviews, we might discern that 80% of those reviews are positive and the average rating of those reviews is 4.0 stars. However, this method could be improved: a positive review mentioning price is not necessarily positive about price. For example:

The food was awesome, and the service absolutely excellent. The price was very high for a coffee-shop style restaurant.

This 5 star review is obviously negative about the price of the restaurant. We need a model that tells us the local sentiment of a sentence or a subsentence in order to be able to understand what elements drive the rating of the review. I’ll explain some of the techniques we have studied, implemented and benchmarked in order to build our Sentiment Mining Tool.

Naive Bayes Classifier

Naive Bayes is the first and the easiest method to classify sentiment in a text. It’s based on the Bayes formula for conditional probabilities:

Bayes Formula



We’ll represent a text by a Bag of Words, which is a set of features “the word w appears f times” for each word w in the sentence and f, the frequency of w in the sentence. Assuming the Naive Bayes assumption that these features are independent, this formula helps us deduce the probability that the sentence is positive (A) knowing that w appears f times (B) for every w. In fact, we can deduce from the frequencies in a large enough dataset the probability for a sentence to be positive (A), and the probabilities of every feature and then of their intersection (B). Training the model on a training set of 10,000 annotated sentences, we get a set of informative features that are helpful to predict whether a sentence is positive or negative. Here are the 10 most informative features we get:

Naive Bayes sentiment-bearing keywords

Naive Bayes classifier’s informative features

This method is the easiest to implement and the big advantage is that it’s completely transparent. When we process it, we know that the classifier found a set of strongly positive or of strongly negative words, and that it is why we classified the sentence in such a way.

How to improve it

However, there are several drawbacks using this method.

First, it fails to identify the neutral class. As a matter of fact, words can have a positive or a negative meaning (“good”, “awesome”, ”horrible”, …) but no word has a neutral connotation. Often, it’s all about the absence of such positively or negatively meaningful words or about the structure of the sentence that reflects the absence of strong emotion. The Bag of Words representation doesn’t address this problem.

It also fails to understand intensity and negations. Comparing “good” and “quite good” for instance, the first one is more likely to appear in a positive sentence than the second one. We tried some methods to address this: adding a list of meaningful bigrams (which mean that we would read “quite good” as a single word for instance), or training the model on bigrams instead of training it on single words, but both didn’t improve our model very much. We also fail to identify negations most of the time, because this model doesn’t take the word order into account.

Most of all, the Naive Bayes model doesn’t perform very well in solving the local sentiment analysis problem. In a long text, having a high frequency of positive words: “sensational”, “tasty”, … makes it very likely that the author is expressing positive sentiment. But as our goal is to determine the local sentiment, we want to process the tool on short sentences and subsentences. (We already have a star rating that tells us the author’s overall sentiment.) We don’t have enough words in the sentence to aggregate so we need to understand very precisely the semantic structure.

The Bag of Words representation is a very bad way to do this. For instance, the sentence “The food could have been more tasty.”, we detect the word “tasty” that is related to a positive feeling, but we don’t understand that “could have been more” is a kind of negation or nuance. Many short sentences are like that, and looking at only a small sentence dataset reduced our accuracy from around 77% to less than 65%.

Rule-based sentiment models

To improve the Naive Bayes methods and make it fit the short sentences sentiment analysis challenge, we added some rules to take into account negations, intensity markers (“more”, “extremely”, “absolutely”, “the most”, …), nuance, and other semantic structures that appear very often near sentimental phrases and change their meanings. For instance, in “The food wasn’t very tasty”, we want to understand that “not very tasty” is less negative than “not tasty” or “not tasty at all”.

We leveraged the results of the Naive Bayes training to build a large vocabulary of positive and negative words. When we process a given sentence, we attribute every word a positive and a negative score, and calculate the overall scores by a precise analysis of the semantical structure based on the open-source library spacy’s pipelines for part-of-speech tagging and dependency parsing. We get a metric for positive, negative and neutral scores, the neutral score being defined as the proportion of words that are neither positive nor negative in the sentence. We used a deep-learning technique to deduce from our training set the relation between these scores and the sentiment. Here are the graphs we obtained for negative, neutral and positive sentences:

Sentiment scores for negative sentencesSentiment scores for neutral sentencesSentiment scores for positive sentences

The model helps us decide very well whether an expressive sentence is positive or negative (we get around 75% accuracy), but struggles understanding a criteria for neutrality or absence of sentiment (on our test-set, it’s wrong 80% of the time). It’s much better than the Naive Bayes, but 75% is less than the state-of-art for positive/negative decision.