Back to top

Enterprise Search Trends: What's to Come in 2016?

Top 4 Impactful Search Features to Watch

Konrad Holl
Konrad Holl
Functional & Industry Analytics Manager

Since facets, auto-completion, and ‘did-you-mean’ features were introduced to search engines, only small changes have taken place over the past few years from a user interface point of view. And although future changes will take place ‘under the radar,’ they will all lead to improved search quality for a better user experience and improved satisfaction. 

As we're greeting 2016, let’s take a look at some of the current and future outlooks for these ‘under-the-radar’ enterprise search features, and what impacts they will have in store for your company's search.


1.  “The Number Games” - Search Accuracy with Engine Scoring

Current methods of evaluating search accuracy are usually based on golden query sets, key documents, zero result searches, or top results. A more sophisticated and automated approach is using log file information from the search engine (search queries) and web servers (clicks) to create a holistic view on search accuracy instead of focusing on limited amounts of information. It’s also possible to review searches on a per user or session basis, taking into account the experience of all users.

A search engine score can then be calculated on the data derived from the log files. Based on this information, searches for a given period of time, user groups, or query types may be replayed with varying query templates to optimize the engine score, i.e. search accuracy.

Below is example of an engine scoring chart resulted from nearly 100 experiments we ran. Watch how we computed engine scores using big data predictive analytics. 


Search engine relevancy scoring chart

Future outlook: Companies may choose to introduce a process for search relevancy tuning, perhaps monthly, where search queries and click logs could be used to compute the current search engine score. Based on that and previous scores, improvements to the query building process can be developed, tested, and deployed to the production environment to optimize the user experience.


2.  “For Your Eyes Only” - Search Result Personalization

Search result quality may also be improved by tailoring the search results to individual users or user groups. With log analytics, users may be clustered based on their search behavior, documents downloaded, user data (profiles), or content contributions. Indicators for tuning these individual queries and results include document popularity, semantics (differing word meanings per user group), or geographic location. This process is completely transparent to the end user and is already used by the Google internet search today (if the user is logged in).

Future outlook: Users browsing on any website or internal portals should be presented with personalized results based on their profile and research history.


3.  “Context is King” - Semantic Search

Current search engines rely on matching words, one to one, with the exception of stop-words and some linguistic base form reduction. However, with increasing computing power, it is now possible to effectively extract entities (statistical phrases or acronyms), compute semantic co-occurrence (semantically related terms occurring together), and perform semantic weighting (based on the document corpus, the user’s history, or both).  

In addition to semantic analysis, a relationship analysis may be performed, extracting semantic triples from plain text. 

From a search perspective, queries may be augmented using search phrases and/or alternatives. Query chains using various methods may be used to improve the search results before returning them to the user. Ranking may also be influenced by semantic criteria. It will also be possible to disambiguate terms in order to match the users experience better (in conjunction with personalization).

Future outlook: The document processing component may be extended to extract semantic information during indexing, possibly using external services like OpenCalais. On the search side, the query processor component may also be modified/tuned to make use of the semantic information.


4.  “The Mind of the Machine” - Deep Learning

Another trend  currently being tackled by Google and Baidu is deep learning. As already demonstrated by IBM with their Watson engine, this is all about ingesting external knowledge and using this knowledge during document processing and query building. Ideally, this external knowledge is already available as information triples or it is generated from data sources like Wikipedia.

Future outlook: Companies may benefit from this technology by generating information triplets from all indexed content, also using public dictionaries like EuroVoc or metadata repositories like the European Union Metadata Repository (MDR).

Evaluating enteprise search options or improving your current search application? Download our free e-book to read about the top 10 features of a high-performing search engine.  

-- Konrad