Enterprise Search Engines are keyword-based and don’t understand user queries or the document semantically. Hence, ends up with 100s of links, not answers..
Enterprise Search Engines treat each user query as different. It neither stores the context nor asks a follow up question to find answer.
Cognitive search employs machine learning to continuously learn and improve answers based on user feedback.
Search engines don’t give an answer but redirect users to a page — a bad user experience.
Search doesn’t integrate into multiple document types, unstructured data, and customer experience applications like Live Chat, Lead Management, Surveys, Recommendations, etc.
Search doesn’t lend itself to a voice based conversational experience.
Avaamo Cognitive Search™ uses our proprietary NLU technology, which has been proven battle-tested and deployed at scale, with millions of users worldwide. Avaamo’s NLU engine provides a core infrastructure already known in the market for its scalability, reliability, and accuracy. In addition, it has the ability to handle many prioritized inference tasks (co-referencing, negations, time reasoning, etc.) on the incoming user query. Our NLU technology offers advanced techniques in intent matching and disambiguation. It has a superior language-understanding ability to handle wide variances in user language such as identifying “bag of words”, domain syntax, misspelled words, and short form phrases.
In addition to our proprietary NLU, our cognitive search leverages BERT technology (Bidirectional Encoder Representations from Transformers) as the base model, which provides the most advanced machine interpretability of language semantically. By implementing the BERT framework, we are able to achieve an unprecedented level of language understanding without human training, to retrieve relevant answers from a large corpus of text.
Context awareness ensures relevant and precise answers. This entails semantic understanding of the user’s question and the content in context to the user’s query. Avaamo Cognitive Search™ personalizes responses based on user’s location, role, and access, giving a precise answer rather than serving up links to multiple documents that cite policies or guidelines.
Specificity: Our approach to performing QA from a large set of documents is to retrieve relevant context out of underlying documents that may contain the answer. This reduced context is input to the reading comprehension model to fetch the final answer for the user query. To retrieve context from a question, we parse the uploaded content semantically and use various models to ascertain specificity, coupled with independent n-gram scoring and fetch the answer in context, based on semantic relevance of the user query.
Improving Accuracy: Signals including identity, location, previous answers—all multiple modes of answering—are combined and ranked to provide unambiguous results. This includes boosting probabilities based on linguistic post-processing of previous response in relation to the question type. Scores at each level of all probable answer candidates are browsed to suggest likely answers based on various hard-lines and then returned with a normalized confidence score.
While the retrieve-read approach is quite powerful, addressing queries that involves multi-step reasoning requires better insight into the specificity of the question. Avaamo Cognitive Search™ queries the user in a human-like manner to get clarification and narrow down the specificity before a question is answered. To handle complex queries, a knowledge graph is dynamically built around documents, support questions, and tickets to facilitate multi-step reasoning. This knowledge graph is created based on pre-built classification models to identify implicit vs. explicit relationships. The knowledge graph is built in real time based on the content that is available, in order to answer queries with a level of specificity that is not available elsewhere in the market.