In enterprise search, context refers to the relevant background information and framework conditions that enable the enterprise search system to interpret search queries more precisely and in a more targeted manner. This includes two levels in particular:
1.Who is searching?
Context makes it possible to customize the search results to the specific user and their role in the company. Context can be anything, starting with the position in the company. For example, an employee from accounting department needs different information than someone from sales department. Context such as name, department and project history can be used to filter results so that only relevant information that matches the user is displayed.
2. What is the searcher's intention?
Context also helps to understand the search intent more precisely. For example, if an employee searches for ‘contract’, they could mean a supplier contract, a customer contract or an employment contract. The company search uses other contextual clues, such as recently edited documents or the employee's current project, to display the appropriate results.
It also considers how the search query is made. In response to the request ‘Give me our sales presentation on Retrieval Augmented Generation from last week’, the system needs to know the date of the presentation and relate it to the date of the enquiry. The system interprets the term ‘presentation’ to search for Power Point or PDF files. Finally, the system interprets ‘Our presentation’ as a presentation from the company/organization the searcher works at.
An intelligent enterprise search such as iFinder has long been able to provide context by extracting metadata from a file, e.g. author, date, file format, file size, etc.
A good search also provides context via the thesaurus function. A thesaurus is an organized compilation of terms from a specialty. This means that synonyms as well as superordinate, subordinate and related terms or abbreviations are cross-linked. If a user searches for ‘GDP’, they find results for gross domestic product or if they search for ‘SUV’, they also get results for off-road vehicles and specific models. Intelligent linguistic functions recognize compound terms such as ‘moonlight’, singular and plural forms and proper names (the name ‘Black’ or the colour ‘black’).
This thesaurus-based query expansion and enrichment (automatic enrichment of documents with metadata) allows to add further context to the search query and thus further improve the accuracy and relevance of the search results.
By using large language models (LLMs), vector searches have recently been added to the full-text search with intelligent linguistics. This algorithm converts texts into mathematical vectors that capture the semantics of words and sentences. It is possible to find information with similar content by not only searching for exact keywords, but also taking into account the context and meaning. In response to the question ‘How can I connect my company laptop to the WLAN?’, the searcher receives the document ‘Instructions for setting up a wireless network connection on your work device.’
Additional context also significantly improves the quality of the results for generative AI. The better it understands who is making a query, what the user has previously looked at in their history and what the intention of the question is, the better the AI's answers will fit. Context has an effect:
1. More relevant answers: context provides the generative AI with additional information that helps it to better understand the meaning of a query and provide an answer that meets expectations.
2. Error avoidance: Context helps to avoid misunderstandings that could lead to incorrect or confusing answers. For example, the same terms can have different meanings in different contexts.
3. Personalization: Context enables AI to respond to the specific needs and requirements of a user, making the results more tailored and useful. For example, should AI respond to an intranet enquiry in a short and fact-oriented manner or create very polite formulations for a customer portal.
Conclusion
The combination of full-text search, intelligent linguistic analysis and vector search creates a powerful system that delivers precise keyword hits as well as similar information in terms of content and takes the searcher and their intentions into account. In this way, we enable a precise and contextualized search that delivers truly relevant information and answers.
