Architecture


Home Prev	Chapter 20. 3.1.0 New Features	Next

Sample Data Includes Surveys

The Concerto sample data set now includes patron surveys, questions, answers, and responses.

Virtual Index Definitions

The practical purpose of Virtual Index Definitions is to supply an Evergreen administrator with the ability to control the weighting and field inclusion of values in the general keyword index, commonly referred to as "the blob," without requiring tricky configuration that has subtle semantics, an over-abundance of index definitions which can slow search generally, or the need to reingest all records on a regular basis as experiments are performed and the configuration refined. Significant results of recasting keyword indexes as a set of one or more Virtual Index Definitions will be simpler search configuration management, faster search speed overall, and more practical reconfiguration and adjustment as needed.

Previously, in order to provide field-specific weighting to keyword matches against titles or authors, an administrator must duplicate many other index definitions and supply overriding weights to those duplicates. This not only complicates configuration, but slows down record ingest as well as search. It is also fairly ineffective at achieving the goal of weighted keyword fields. Virtual Index Definitions will substantially alleviate the need for these workarounds and their consequences.

A Virtual Index Definition does not require any configuration for extracting bibliographic data from records, but instead can become a sink for data collected by other index definitions, which is then colocated together to supply a search target made up of the separately extracted data. Virtual Index Definitions are effectively treated as aggregate definitions, matching across all values extracted from constituent non-virtual index definitions. They can further make use of the Combined class functionality to colocate all values in a class together for matching even across virtual fields.
Configuration allows for weighting of constituent index definitions that participate in a Virtual Index Definition. This weighting is separate from the weighting supplied when the index definition itself is a search target.
The Evergreen QueryParser driver returns the list of fields actually searched using every user-supplied term set, including constituent expansion when a Virtual Index Definition is searched. In particular, this will facilitate Search Term Highlighting described below.
Stock configuration changes make use of pre-existing, non-virtual index definitions mapped to new a Virtual Index Definition that implements the functionality provided by the keyword|keyword index definition. The keyword|keyword definition is left in place for the time being, until more data can be gathered about the real-world effect of removing it entirely and replacing it with Virtual Index Definition mappings.
New system administration functions will be created to facilitate modification of Virtual Index Definition mapping, avoiding the need for a full reingest when existing index definitions are added or removed from a virtual field.

Increased use of Metabib Display Fields

We use Metabib Display Fields (newly available in 3.0) to render catalog search results, intermediate metarecord results, and record detail pages. This requires the addition of several new Metabib Display Field definitions, as well as Perl services to gather and render the data.

We also use more Metabib Display Fields in the client. As a result, bibliographic fields will display in proper case in more client interfaces and in Evergreen reports.

Interfaces

A new AngularJS "MARC Search/Facet Fields" interface has been created to replace the Dojo version, and both have been extended to support Virtual Index Definition data supplier mapping and weighting.

Settings & Permissions

The new Virtual Index Definition data supplier mapping table, config.metabib_field_virtual_map, requires the same permissions as the MARC Search/Facet Fields interface: CREATE_METABIB_FIELD, UPDATE_METABIB_FIELD, DELETE_METABIB_FIELD, or ADMIN_METABIB_FIELD for all actions

Backend

There now exist several new database tables and functions primarily in support of search highlighting. Additionally, the QueryParser driver for Evergreen has been augmented to be able to return a data structure describing how the search was performed, in a way that allows a separate support API to gather a highlighted version of the Display Field data for a given record.

Default Weights

By default, the following fields will be weighted more heavily in keyword searches. Administrators can change these defaults by changing the values in the "All searchable fields" virtual index in the "MARC Search/Facet Fields" interface.

Title proper
Main title (a new index limited to the words in the 245a)
Personal author
All subjects

In addition, note indexes and the physical description index will receive less weight in default keyword searches.

Re-ingest or Indexing Dependencies

With the addition and modification of many Index Definitions, a full reingest is recommended. However, search will continue to work as it did previously for those records that have not yet been reingested. Therefore a slow, rolling reingest is recommended.

Performance Implications or Concerns

Because the Metabib Display Fields infrastructure will eventually replace functionality that is significantly more CPU-intensive in the various forms of XML parsing, XSLT transformation, XPath calculation, and Metabib Virtual Record construction, it is expected that the overall CPU load will be reduced by this development, and ideally the overall time required to perform and render a search will likewise drop. It is unlikely that the speed increase will be visible to users on a per-search basis, but that search in aggregate will become a smaller consumer of resources.