Editor' note: Dr. Sherry Mead is a cognitive psychologist with expertise in the user experience of search features and functions. This blog is the second of a two-part series on the latest in user interface design for search functionality.
Sherry Mead, User Experience Architect
Identifying and Implementing Facets
In Part 1 of this blog post, I explained how facets can be a very useful feature to enterprise users for the search function. In this part, I will describe how to identify and implement the facets. Fortunately, structured data in databases is already "tagged" with facet values because attributes with the following characteristics are good candidates for use as facets: Much value can be added to enterprise search applications by enabling users to browse unstructured data, such as e-mails, full-text files, and Web pages, using facets. The challenge is to add structure to the unstructured data. Requiring users to manually tag documents with predefined facet values (a controlled vocabulary based on business objects and their attributes) would be laborious as well as error-prone. Using optional, freely assigned tags as facets is powerful and useful, but can produce idiosyncratic and inconsistent results. Named Entity Recognition (NER) systems represent another possible option for adding structure to unstructured data. An automated system would dramatically reduce user effort and allow more consistent application of a predefined or discovered set of tags. Clustering algorithms generate tags on-the-fly. However, it would be advantageous to use the structure that already exists in enterprise application databases to surface relationships among structured and unstructured data in both the search interface and the enterprise application interface. Users will inevitably add information to data that does not exist in structured data sources so that a combination of free tagging, clustering, and the addition of predefined structure is likely to produce the best results. Flat or Hierarchical Facets A defining characteristic of facets is that they are independent of one another. However, designers have at least two options when presenting hierarchical facets to users. They can surface the hierarchy via tree or drill-down interfaces, or they can present different levels of a single hierarchy as if they were separate facets. Hearst (2006) has studied the latter presentation and determined that is it usable when the semantic (meaning) relationship among the hierarchy levels is clear to users. If enterprise application users are familiar with business intelligence dimensions, tree or drill-down facets might be more appropriate. Business Intelligence (BI) dimensions allow users to view data by categories, such as Geography, Time, and Product. BI dimensions and facets have a lot in common. Facets and dimensions must be independent. Their value sets comprise discrete values or ranges. They can be flat or hierarchical, and both types of entities are used to select and filter data. They differ in that dimensions are used to aggregate numeric data, while facets are used to select characteristics of documents or objects. Conclusion In summary, faceted search uses concepts that business users are likely to be familiar with in support of an easy- and pleasant-to-use search and browse interface. Faceted search has numerous advantages over keyword-only and advanced fielded search user interfaces. Further, structured data in database records and business intelligence dimensions give enterprise search application developers an advantage in designing and implementing facets.
