Part 2. NER for all tastes: extracting information from cooking recipes
In the previous articles, we constructed two label datasets to train machine learning models and develop systems able to interpret cooking recipes.
This post dives into the extractor system, a system able to extract ingredients, quantities, time of preparation, and other useful information from recipes. To develop the service, we tried different Named Entity Recognition (NER) approaches.
Hold on! What is NER?
ER is a two-step process consisting of a) identifying entities (a token or a group of tokens) in documents and b) categorizing them into some predetermined categories such as Person, City, Company... For the task, we created our own categories, which are INGREDIENT, QUANTIFIER and UNIT.
NER is a very useful NLP application to group and categorize a great amount of data which share similarities and relevance. For this, it can be applied to many business use cases like Human resources, Customer support,Search and recommendation engines,Content classification, and much more.
NER for the Smart Recipe Project
For the Smart Recipe Project, we trained four models: a CRF model, a BiLSTM model, a combination of the previous two (BiLSTM-CRF) and the NER Flair NLP model.
CRF model
Linear-chain Conditional Random Fields - (https://medium.com/ml2vec/overview-of-conditional-random-fields-68a2a20fa541)[CRFs] - are a very popular way to control sequence prediction. CRFs are discriminative models able to solve some shortcomings of the generative counterpart. Indeed while an HHM output is modeled on the joint probability distribution, a CRF output is computed on the conditional probability distribution.
In poor words, while a generative classifier tries to learn how the data was generated, a discriminative one tries to model just observing the data
In addition to this, CRFs take into account the features of the current and previous labels in sequence. This increases the amount of information the model can rely on to make a good prediction.
Fig.1 CRF Network
For the task, we used the Stanford NER algorithm, which is an implementation of a CRF classifier. This model outperforms the other models in accuracy, though it cannot understand the context of the forward labels (a pivotal feature for sequential tasks like NER) and requires extra feature engineering.
BiLSTM with character embeddings
Going neural... we trained a Long Short-Term Memory (LSTM) model. LSTM networks are a type of Recurrent Neural Networks (RNNs), except that the hidden layer updates are replaced by purpose-built memory cells. As a result, they find and exploit better long-range dependencies in the data.
To benefit from both past and future context, we used a bidirectional LSTM model (BiLSTM), which processes the text in two directions: both forward (left to right) and backward (right to left). This allows the model to uncover more patterns as the amount of input information increases.
Fig.2 BiLSTM architecture
Moreover, we incorporated character-based word representation as the input of the model. Character-level representation exploits explicit sub-word-level information, infers features for unseen words and shares information of morpheme-level regularities.
NER Flair NLP
This model belongs to the (https://github.com/flairNLP/flair)[Flair NLP library] developed and open-sourced by (https://research.zalando.com/)[Zalando Research]. The strength of the model lies in a) the use of state-of-the-art character, word and context string embeddings (like (https://nlp.stanford.edu/projects/glove/)[GloVe], (https://arxiv.org/abs/1810.04805)[BERT], (https://arxiv.org/pdf/1802.05365.pdf)[ELMo]...), b) the possibility to easier combine these embeddings.
In particular, (https://www.aclweb.org/anthology/C18-1139/)[Contextual string embedding] helps to contextualize words producing different embeddings for polysemous words (same words with different meanings):
Fig.3 Context String Embedding network
BiLSTM-CRF
Last but not least, we tried a hybrid approach. We added a layer of CRF to a BiLSTM model. The advantages (well explained here) of such a combo is that this model can efficiently use both 1) past and future input features, thanks to the bidirectional LSTM component, and 2) sentence level tag information, thanks to a CRF layer. The role of the last layer is to impose some other constraints on the final output.
Fig. 4 BiLSTM-CRF: general architecture
What about performance?
(https://medium.com/@condenastitaly/when-food-meets-ai-the-smart-recipe-project-8dd1f5e727b5)[Read the complete article on medium], to discover that and more about this step of the Smart Recipe Project.
When Food meets AI: the Smart Recipe Project
a series of 6 amazing articles
Table of content
Part 1: Cleaning and manipulating food data
Part 1: A smart method for tagging your datasets
Part 2: NER for all tastes: extracting information from cooking recipes
Part 2: Neither fish nor fowl? Classify it with the Smart Ingredient Classifier
Part 3: FoodGraph: a graph database to connect recipes and food data
Part 3. FoodGraph: Loading data and Querying the graph with SPARQL
Top comments (0)