Context based Search - Is it the future of Search technology?
Many studies and analysts have quoted that, white collar workers spend 40% of their time searching for, analyzing and assembling information which makes the process incredibly inefficient. They first have to locate the information in disparate systems, analyze it using various application interfaces and then compile the relevant and related pieces out of all the information collated. Search platforms hold the key to integrating this workflow and delivering new high performance tools, that provides users with a unified and actionable view of information. Search technology has undergone a major transformation from a simple text box producing a set of matching results and is now a technology much beyond just indexing and serving the document content based on the search term. Search domain is now being termed as "Information Access and Retrieval" in the overall Information management space.
Search no longer is just a connection between the search keyword and the indexed content, but is more of applying the user context along with the search keyword to produce more meaningful results based on the users intent of searching. When the search giant Google is asked about the future of search, they mention it as "context based search" and not just query based. The way Google sees the future of search, is its place as your "assistant" while you navigate the web. Put differently, the search platform has become a broker between the user context and the available data and applications. This role requires the search platform to understand the user's context in order to give precise answers--a task that requires "Contextual Search".
Some latest adoption trends for the information access technology are:
1. Portal UI mash ups and Knowledge Management solutions where search technology connects and integrates people and information within the enterprise or the web.
2. Search over BI and BI over Search - Search can augment conventional BI by tapping into new sources of unstructured data and enriching the analysis through seamless data navigation and filtering the relevant information.
4. Competitive/Corporate Intelligence for the extraction of unstructured content within the enterprise and the web.
5. Entity extraction and Sentiment Analysis using Ontologies and Semantic intelligence processing. Please refer my earlier blog on how search helps in decision intelligence to understand points 4 and 5..
Common to all these latest trends is the search platform's role in providing the users with precise situation-specific information and functionality. In this blog I will focus on the search platform's role as a broker between the user context and the available data and applications, and how contextual search can be applied by capturing the front end interactions and the end user context in order to deliver relevant answers or relevant search content. Search platforms and vendors have ensured that part of the contextual search gets built in the core engine itself. Most of the commercial and open source search tools will have features like synonyms, misspelled queries or did you mean, auto suggestions, soundex search, faceted search , drill down navigations, content spotlighting, keyword redirects, rule based merchandizing already built in as a part of the core system.
First, let me explain some of these search platform backend features by giving few examples..
1. To improve the search precision, search will rate the documents closer to where you are present in the site hierarchy. This is very much possible through faceted search and the drill down navigations supported by the search system. Search systems have inherent capabilities to integrate with the organization ontology and folksonomies, making the navigation and dynamic faceting relatively easy and providing a rich customer experience. For e.g. In a retail site, if the user is in CD/DVD category and if he types the search keyword as "Spiderman", he would be presented with Spiderman CD's and DVD's rather than Spiderman outfits and masks.
2. System will provide automatic suggestions to the search keywords when user starts typing in the search text box. Again this is achieved through the backend functionality where it comes up with keyword suggestions based on the data present within the search index. This is exactly similar to what Google provides in its search box when users starts typing their search terms.
3. The system will automatically correct the search term in case if you have misspelt it and will by default produce the matching results based on the auto corrected search term. Platforms also give you option of Did you mean incase if you are not sure about the search term you want to execute. For e.g. if you search for bear when you are looking out for beer. Few platforms also have soundex feature which enables to retrieve the results in case if you are not very sure about the spelling of the keyword.. For example if you search for Filipines it will still retrieve correct results assuming the search keyword as Philippines.
4. Business users can configure one way and two way synonyms in the system based on the data and their industry domain. For e.g. people searching HP are automatically given results for Hewlett Packard and HP.
5. Content spotlighting or Rules Manager functionality provided by the search vendors is the latest trend to ensure that the content is manually overridden or promoted based on the search terms, navigation clicks, path in the taxonomy or based on the user profiles. In relation to the earlier example of the search keyword Spiderman, you could use content spotlighting to promote/cross sell Spiderman outfits or toys when user searches for Spiderman in the CD/DVD category of the retail site.
6. Someone searching for "Customer Service" in a retail store should not be shown matching results related to it, but should be directly redirected to the Customer Service home page of the site. Keyword redirect feature of the search tool helps the site achieve this functionality and is again a widely used configuration within the business users.
For each new demand, a new platform functionality was added to alleviate the problem, always making advancements in the search technology. Contextual search now accelerates the pace of change and the ability of the system to capture user context has the potential of creating a new quantum leap in the industry. The sole purpose of capturing the user context from the front end is to improve the search precision. In order to achieve this, the platform either has to be extended, built or fine-tuned to add the contextual metadata into the search query and have those relations and associations stored in the search index along with the content as metadata, thus giving additional data for the backend to produce more relevant results.
This contextual meta data can be either few or all of the below parameters..
1. User profile and segment information. This data can be extracts from the 360 degree view of the customer data which can be leveraged to produce the best user context search results.
a. Explicit and Implicit interests and preferences
b. Location - Country, State, city, address, current coordinates etc.
c. Age, gender, income range
d. Devices used - web, mobile, POS etc.
e. User browsing and searching history
f. User interactions on the site
2. Navigation path
a. Navigation path taken
b. Current position of the user on the site
3. Entire Search keyword history
4. User ratings and reviews
5. Date and time of search : Festivals - Christmas, Diwali period.. time of the day - morning, afternoon, evening, night etc..
6. Weather in question during the search: Summer season, rains, winter season, heavy snows etc..
7. User Mood: Hungry, Positive and negative Emotions, excitement, angry
Again just passing this as extra meta data along with the search query may not help, unless you have an association of these parameters with the indexed content. Hence it is important that when the content is indexed in the search platform the business ensures that the content is tagged properly based on the above factors. It is more of finding the similarities and links between the front end contextual metadata and the backend content stored with appropriate meta data to boost the relevance. Search platforms should be solutionized to have these as navigational attributes or facets within the system, which will help in faster and efficient retrieval and matching. The search user context component will end up having links between the documents, users, categories, products, date and times, user segments and the contexts of the query will give you valuable information on how to rate and rank this particular information. Next stage would be to render the dynamic results, made as configurable dynamic mash-up presentation of data elements and application components in relation to the end users role, profile and the context.
Let me make my point clear with few existing implementations in the contextual search space..
1. If user searches for the keyword "phone accessories", then present him with iPhone accessories first in the result set as you know that he has already purchased a iPhone from your shop last week.
2. For a news site, if user has previously searched for archived news on "Sachin Tendulkar" and then does his search on "World Cup records", then he should be presented with cricket world cup records and happenings at the top followed by soccer or other sports where world cup also takes place.
3. Take into account the users navigation path in the knowledge management solution. For example. If user has navigated from the HR section in the intranet site and searches for policies, he should be shown the HR policies with a higher relevance than corporate, department or legal policies..
4. Presenting personalized search results, products or promotions on the mobile device based on user's selection. For e.g. Based on the users current coordinates a search service provider can give the search results accordingly considering options to his vicinity.
5. Giving higher preference to Beer outlets/ restaurants if user searches for alcohol hangouts especially if the search is done during Friday afternoon or evening time
6. Apparel store capturing implicit and explicit preferences like fashion preferences, presenting the display, promotions, advertisements and search results based on users interests, age and income range.
With this technology, we expect the emergence of innovative and search-driven end-user applications and already see the adoption of this technology in large consumer portals, ecommerce shops, knowledge management solutions, Search Analytics and decision intelligence solutions and expect this to become prevalent in enterprise search also.. In future your query and your decision as to what to click upon, as well as your follow up searches may become part of the statistical "contextual click model" developed around future searches by others for the same query. Your results may be modified based upon such a model, but only for a limited query session.