Artificial intelligence is a competitive field. At Nowigence, our growing technology organization contains an international team of professionals. Our team works in a fast-paced development environment. Most of them have a master’s degree in the field of computer science or data analytics from one of the top universities in the US. To stay on top of the competition, we are a learning organization that works on shared visions to seek continuous improvement and product innovation. We also work with research universities such as George Mason University, the State University of New York at Albany (SUNY), and Rensselaer Polytechnic Institute (RPI) on a regular basis to foster novel and empirical NLP / machine learning methods and applications to both academia and industry.
Nowigence utilizes machine learning in the following natural language processing (NLP) tasks. We provide large-scale automatic solution on each task to our subscribers.
- Named-entity recognition (NER) – Business-critical signals are extracted in the form of named-entities from narrative free-texts. Being the key indicators of business events, business-critical signals are commonly found in both internal and external data sources. These signals (e.g., companies, products, person names, temporal expressions) are stored at a client level longitudinally to enrich their institutional memory. We then use this memory to exploit hidden patterns and generate better decision boundaries. We tackle the entity extraction task using a sequence labeling classifier after automatic feature engineering using bi-directional long-short-term memory (bi-LSTM) with word embeddings. Our NER method is state-of-the-art when evaluated on the publicly-shared ground truths.
- Machine comprehension – Our platform comprehends natural language queries (questions) and extracts the most probable answers from both internal and external data sources (user’s institutional memory). We are capable of handling automatic machine comprehension at a large scale in real-time using bi-directional attention flow. This enables us to provide specific answers immediately after users have uploaded their own data such as news articles, internal emails, and meeting notes to our Pluaris platform. To complement the answers, we also provide related articles and summaries of business entities on the topics of natural language queries. These entities are extracted using the NER component mentioned above.
- Information retrieval – Our platform filters news articles as well as other business documents based on specific companies, industries, and products we monitor on a daily basis. We have a machine learning classifier in place to identify the information that is relevant to the users. We estimate the relevancy between monitored topics and business events using linguistic features of each news article derived from rigorous feature engineering. For better user experience, we also conduct near-duplicate detection to reduce the number of redundant news articles by calculating the distances between their digital fingerprints.
- Auto-summarization – We automatically generate key points and main drivers for each business article. Key points make our users understand the main theme of each article without reading its full content. Main drivers demonstrate the causal factors of business events mentioned in the article. We dynamically process each article on a sentence-level to find the most valuable and informative utterances and rank them in order. Co-reference resolution is also applied in order to replace pronouns with their corresponding co-referents (entities) for higher readability.
In addition to these major components, our platform is also capable of (1) sentiment analysis to detect the polarity of news articles and user-uploaded documents to pinpoint risks and opportunities; (2) multi-label classification to project both internal and external articles and main drivers to our proprietary business categories; and (3) active learning to iteratively improve prediction accuracy based on user feedback.
In addition to the novel machine learning algorithms and NLP techniques, our AI platform is very user-centric. We give a high level of control to the users. Our objective on product development is to get away from “yet another impractical AI platform for a technological vision showcase”. We believe the benefit of artificial intelligence is to make our users’ life easier and enable them to work more effectively. We encourage our platform subscribers to accept and embrace technology by giving them control of these AI components. Unlike many of our competitors, we allow users to bring their own data in various formats, creating an institutional memory that belongs only to their marketplace. Every AI component mentioned in this article can be applied to the institutional memory.
Interested in what we do? Schedule a quick product demo here! We will show you these components in real-time.