Natural Language Processing Parsing software to achieve high accuracy of an ATS

computer vision

Client

A company that builds talent matching algorithms for enterprises, recruiting software vendors, career sites, and staffing agencies.

Problem

Low accuracy rates of open-source NLP parsers. The client was facing challenges gaining new customers and growing market share because their solution was unable to compete with larger ATS.

Solution

We developed a neural network and augmented it with NLP to process the input text and classify every word (a NER system). The training dataset was small, so we used unsupervised learning (e.g. Word2Vec) to achieve accurate results.

Data Scientists at BroutonLab have split the project into two stages:

Stage 1

Research and Development. We trained several deep neural networks to retrieve structured information from text automatically. The networks performed a semantic search, topic modeling, and text matching.

Unsupervised learning algorithms that we used helped our client save money on manual text labeling. The algorithms are domain/ language independent.

We trained custom word embedding (word2vec, glove vectors, Google transformer/BERT) to increase the accuracy of the models.

We increased the diversity of training samples and decreased over-fitting by using data augmentation techniques.

Technologies used

Python, Keras/Tensorflow, Pytorch

Stage 2

Porting and deployment of models in the server.

With Tensorflow serving, we trained neural networks as individual micro-services scalably. We developed an API with Goland and gained a top speed of processing user queries.

We used binary GRPC protocol to perform the communication between Goland and Tensorflow serving.

Get a Free Consultation Now