Five Levels of Design Considerations for Effective Search Implementation
Author: Brett Matson, Managing Director, Funnelback
These days there are few knowledge workers who fail to recognise the benefits of highly effective search. This article outlines five levels of search implementation design considerations that progressively lead to more effective search across web sites, intranets and other corporate repositories.
Level 1: Out-of-the-Box
This is what search in the 1990s was all about. It’s quick and easy and sometimes it works. It’s also the level of sophistication achieved by global search engines such as Google that have to deal with indexing millions of heterogeneous web sites. At this level, the search engine is not aware of what it’s crawling, it just attempts to crawl the content systematically (e.g. by following hyperlinks). There is no attempt at integration of the search engine and the content server. Search at this level can be hit and miss due to the lack of context for the ranking system to draw from.
Design considerations at this level:
- What is the scope of the content to be crawled?
- How often should the content be crawled?
Benefits achieved by this level:
- Full text search
- Basic level of dynamic contextual navigation.
Level 2: Repository Awareness
At level 2, the search engine is provided with context about the repository that it’s crawling. This can lead to a wide variety of measurable benefits to the effectiveness of the search implementation.
Design considerations at this level:
- Ranking: Often this involves adjusting the ranking to bias away from characteristics of the repository that result in sub-optimal ranking. For example, a search engine that indexes a web site with no clear site structure should be configured to bias this characteristic out of the ranking algorithm in favour of more discriminating characteristics such metadata, link information or recency.
- Content update feeds: Can the repository inform the search engine when updates have occurred so that indexing of new and changed content can take place in a timelier manner?
- Document level security: Should the search engine restrict access to search results from some users?
- External metadata: Are there external sources of useful metadata (e.g. not published within the content) that could be provided to the search engine?
- Contextual de-duplication: Can the search engine be provided with clues as to which content pages are duplicates so as to provide more effective de-duplication of search results.
Benefits achieved by this level:
- Adherence to access controls
- More frequent index updates
- Indexing of external metadata
- More effective ranking of search results
Level 3: Content Awareness
At this level the search engine understands the context of not only the repository it’s accessing but also the content it’s indexing.
Design considerations at this level:
- Faceted navigation: Is there an existing metadata taxonomy available that could be used to drive faceted navigation of the content? This taxonomy can be as complicated as a hierarchical pharmaceutical database or as simple as having a neatly organised URL structure and date metadata for each page. Faceted navigation provides a highly effective way of filtering search results, particularly in applications where there is limited link information to assist the ranking algorithm.
- Use of metadata: In addition to faceted navigation, metadata can be used to enable fielded search in advanced search forms and also for the purpose of displaying customised search result summaries. Metadata can also factor into the ranking (e.g. which fields should be indexed? What is the relative weight of each field in the ranking? How should content recency bias the ranking?)
- Adding structure to unstructured content: Structured metadata can be beneficial to search effectiveness for all the reasons mentioned above. For content that has inadequate metadata, a search engine that understands the visual design of the content can often scan the content looking for metadata, extract it out and then incorporate it back into the index as structured metadata so as to provide the full range of benefits from metadata.
- Targeted indexing of content fragments: Can the indexer be configured to skip low quality supplementary information (e.g. site navigation, headers, lists of news headlines, horoscopes, etc.). This information can often cause a flood of false matches in the search results and also low quality search summaries in highly relevant search results.
- Improving search result titles: Often the search result titles displayed by an out-of-the-box search engine are sub-optimal. These titles can be improved by having the search engine know where to find higher quality titles (e.g. within the document), by stripping out repeated text in titles (e.g. the organisation name) and adding context to low quality titles (e.g. adding the date published in the case of multiple results having the same title).
Benefits achieved by this level:
- Higher quality ranking
- Faceted navigation
- Fielded search
- Customised search result summaries
- Higher quality contextualised search result summaries and result titles
- Ability to add content structure where it doesn’t exist and benefit from faceted navigation, fielded search and customised result summaries.
Level 4: User Centred Interface Design and Web 2.0
This level is all about designing the search interface and search results pages with specific users in mind. It’s also about having the search engine learn to produce better results by analysing its past interactions with users.
The Search Interface
The user interface design should match the skills of the users and they should assist them in carrying out specialised search tasks. The search engine can assist with this by giving users the tools to construct meaningful queries. Just as importantly, the optimal level of information should then be provided to help the user make informed decisions for refining their query, selecting results to view and suggesting appropriate follow-up action when no results are found.
On-the-fly Personalised Navigation Structure
An effective way of assisting users to refine their query is through the use of an automated sub-topic navigation system. These work by analysing the query and the search results to produce an on-the-fly personalised navigation structure. The advantage of these systems is that it provides a dynamic summary of the search results that often reveals otherwise hidden content, which then ‘cross-sells’ the user to content that may have otherwise been overlooked. An example of this technology is the Flustering engine that forms part of the Funnelback search engine.
Automated Learning from User Behaviour
An effective search technology will also have the means to draw guidance from users. This is often in the form of biasing the search ranking towards more commonly clicked results (known as click feedback) and also in indexing collaborative user tags. Indexing user tags helps to alleviate terminology disconnect between the published content and user queries.
Design considerations at this level:
- Search features: The decision to use features such as automated sub-topic navigation systems, click feedback and user tagging.
- Search driven thesaurus: Given the background of the users, would they benefit from a search driven thesaurus for refining their queries (e.g. broader terms, narrower terms, related terms, etc.)?
- Hard coded search results: Will the users be submitting common queries that would benefit from having hard-coded results appear at the top of the ranking?
- Tailored advanced search forms: Does the nature of the search tasks being performed and the skill level of the users warrant the use of one or more tailored advanced search forms? If so, what kind of advanced search fields would best match the information domain and the types of search tasks being performed?
- Customised summaries: Can the users make use of certain elements of information in the search results to help them quickly evaluate the search result list? Should a reference image be displayed next to each search result? Should the search results be in a table format rather than a standard web search format?
- Follow-up action for when no search results are returned: What advice can the search engine give to users for queries where no results are found? This could be as simple as providing contact information for someone who can assist, or providing a link to search other repositories.
Benefits achieved by this level:
- Empowering users with tools that match their information domain, the nature of their search tasks and their skill level to enable them to make effective use of the search technology
- Higher quality search results via the user of user click feedback and indexing of user tags
- Enabling users to more efficiently scan search result summaries
- Giving users useful advice when common queries are submitted and direction on how to proceed when no search results are found
- Cross-selling users via the use of personalised navigation structures
Level 5: User Context in Search Ranking
At level 5 the search engine can start to get an almost unnerving level of understanding about what kind of search results are going to be useful to individuals or groups of users.
Design considerations at this level:
- How to make effective use the user’s demographic? This could involve up-weighting technical documentation for IT staff, up-weighting marketing material for sales prospects or up-weighting products previously purchased by the search user.
- How to make effective use of geographic proximity? If the location of the user is known (e.g. post code, latitude/longitude, country, regional office, etc.) then this can be used to bias the search results toward content that is geographically closer to the user. Would the user benefit from having the search engine mash-up with an online mapping system so that the search results can be plotted visually?
Benefits achieved by this level:
- More effective search by providing information that is relevant to the demographic and geographic context of the user
What Comes After Level 5?
If you exceed level 5 you risk making the search engine become self-aware. At this point the search engine is less concerned with attempting to answer user queries and is more interested in answering questions about its own existence and the world within which it exists. There is little hope of having a productive search engine once it’s reached level 6, so even with the availability of highly effective search technology, many content publishers prefer to mitigate this risk by only implementing up to level 1. It’s your call.
Recent Solutions have been