1.7 Context Search

How to enable contextual search (where the relevance of retrieved information improves, more the system is aware of the search context) in enterprise applications with large linear and non-linear structured, semi-structured and unstructured data sets?

 1.7.1 Problem

An enterprise has large content and data – structured, semi-structured and unstructured. It has to serve its employees, customers and partners globally with a need to support multiple languages. The users need to be provided most relevant content contextually, correlating all related product, geographical, organizational, temporal and market information as well as previous information searches. The search solution has to account for multiple content sources (structured and unstructured), process existing sources of content, onboard new sources of content, apply appropriate fine-grained security policies while presenting the search results.

 1.7.2 Solution

The solution pattern for Context-aware Search usually involves the following components:

  • Automatic indexing

The goal of automatic indexing is to establish an index for a set of documents that has to facilitate future access to documents and to their content. Usually, an index is composed of a list of descriptors, each of them being associated to a list of documents and/or of parts of documents to which it refers. In addition, theses references may be weighted. When searching to answer the users’ queries, the system looks for a list of answers, of which an index is as close as possible to the demand. As a consequence, indexation could be seen as a required preliminary to intelligent information retrieval, since it pre-structures textual data according to topic, domain, keyword or area of interest.

• Relevancy

Any modern search engine index spans tens or hundreds of billions of pages, and most queries return hundreds of thousands of results. Since the user cannot parse so many documents, it is a pre-requisite to rank results from most relevant to least relevant.

• Conversational User Interface

“Question/Answering” addresses the problem of finding answers to questions posed in natural language; answering is the task which, when given a query in natural language, aims at finding one or more concise answers in the form of sentences or phrases. Contexual Search platform must ‘understand’ conversation, support the addition of new content (and the context it delivers, too), and learn about you from the questions you ask.

• Continuing refinements in deciphering context from evidence

Correctly correlating evidence from structured and unstructured data starts with understanding the context of each data point. Detecting patterns and deciphering meaning involves text analytics, especially with unstructured information. This is achieved using a variety of techniques, including machine learning and statistical, modern analytics learn from feedback, adding cognitive power.

Incorporate text analytics during data ingestion, to perform NLP, to support evidence correlation, and to detect similarities and differences – and your search solution will evolve with trends in users’ questions. Context and learning combine to make answers more accurate and precise

• More explicit linkage between analytics, answers, and decision outcomes

Answers shape decisions. Decisions produce outcomes. Historically, search solutions rarely linked answers and the outcomes they shaped.

Automated relevance tuning, using learning algorithms to analyze outcomes and suggest adjustments, completes the ‘cognitive’ process.

 1.7.3 Application

Contextual Search services build a strong association between questions and immediately useful answers. More frequent use, like practice, develops habits.

Habits, in turn, reflect and drive loyalty.

Just as enterprises seek to attract loyal, repeat customers, Contextual Search rebuilds ‘user equity’ in their applications.

 1.7.4 Examples / Use-cases

A customer searching through a vendor’s ecommerce site looking for a specific product is automatically guided with a filtered choices based on the knowledge of existing versions of related product installed in the customer’s infrastructure, along with a choice of subscription prices to choose from with ready-to-install.