Context-Aware Search System (CASS)

Context-awareness is a concept that makes use of contextual information with respect to the user’s point of view used to make effective use of the current application being operated on by adapting itself to the user’s custom settings and provides a better learning experience both which is personal and context centric that sounds more effective and efficient. The user’s custom settings include information such as user’s location, time, temperature, personal information, role, purpose of use, other users or relevant devices, i.e. proximity, recent activity, sentiment, etc. There are certain issues to take care of like privacy, security and efficient deployment. We do encounter challenges while assembling contextual information.

The contextual information varies from environment to environment and has its own limitations and scope which is very hard to predict. It is very difficult to identify the right parameters to get it working right. This information can be obtained from different sources like web browsers, sensors, cameras, microphones, GPS receivers and others. It makes predictive analytics more powerful and paves a way to develop more sophisticated practical applications that play a major role in fostering consumer-specific ones. Some interesting context-aware applications would be calendar alerts, weather alerts, commute alerts, news information, technology recommendations and updates, personalized home, workspace or school settings of the software applications we use, search engines, etc. For instance, let us consider the case of a Wikipedia knowledge base which is a collection of documents.

Each document contains hyperlinks which outline specific information, inter-wiki links and disambiguation pages which explain possible ambiguous terms. Each document(title) can be treated as an entity, its description page as context and hyperlinks in the page as query terms. A classifier which is responsible for measuring the similarity between each hyperlink’s context and for concluding if a hyperlink and entity should be linked together or not, in other words called entity-linking problem. Having to recommend related entities for a given query is an important feature to be included in search engines.

This can be done with queries with explicit entities, but it is difficult for queries without explicit entities as the meaning of the text cannot be inferred with the models currently in place. The context will definitely play a significant role in entity recommendation highlighting the semantic power of queries.

This can be modelled with the help of attentive deep neural networks. The entity recommendation system comprises of three modules – query processing, candidate generation and ranking. The query processing comprises of three steps – refactoring queries which are a preprocessing step, deriving entities and ideating entities which is the final step. The candidate generation fetches the output of the query processing module and retrieves related entities using heterogeneous graph embedding for unsophisticated queries and a deep collaborative matching model for relatively complicated queries (Sundermann, Domingues, Marcacini, & Rezende, 2015). Context-aware recommender systems (CARS) is yet another fascinating feature to focus on. It helps us predict user preferences and inclinations with the help of available contextual information to improve productivity.

This can be implemented using the LUPI-based Incremental Hierarchical clustering method which takes three kinds of selective information – traditional bag of words (technical information), combining named entities (privileged information I) with domain terms (privileged information II) (Jia, Zhang, & Hua, 2019). Contextual computing can be beneficial to the healthcare and medical industry, military operations, tourism sector and many other areas of application.