The “next-generation” catalogues of academic libraries provide a discovery layer that contains faceted classification and search features and suggested topics selected by their rankings. To improve the discovery process, this paper demonstrates an interactive faceted visualization box, termed VisFacet, that extends the catalog interface and allows users to narrow or broaden their search results filtered by suggested topics or facets in an interactive manner. The VisFacet software visualization module was integrated into and contributed to the VuFind open source software system. VuFind is a development portal that enables libraries to customize their own catalogue interfaces and discovery layer. Thus, extending VuFind with VisFacet provides all catalogues using VuFind at their system's core with the benefit of having an infographic interactive search box. We will describe the challenges encountered during the development of the VisFacet project, including the user discovery satisfaction questionnaire results.
The traditional academic library catalogue enables end users to search for and
find requested information. With the development of the Internet, and the
introduction of Google as a major player in the information retrieval arena, the search
patterns began to change. The Google-like search method influenced the way that
library catalogues are being developed to offer a similar user experience [
The faceted classification feature that classifies items into multiple independent
categorization schemes (or facets) is a topic from library science that has become
popular in information computing [
VuFind (www.vufind.org) is an open source, next-generation library resource portal that enables the development of a customized, faceted navigation system. VuFind provides a full set of modules to produce a customized cataloging system layer in which libraries can implement all the features they want, modify the existing modules to best fit their needs, and add new modules to extend their resource offerings. The Library Authority of the Hebrew University customized the VuFind system modules and built the HUfind next-generation catalogue. HUfind has been in production since 2012. Our VisFacet framework extends the VuFind system’s capabilities with the visualization feature, and all catalogues that use VuFind at their system’s core (such as HUfind) can benefit from VisFacet. Moreover, today there are several development platforms for modern cataloging of library systems, and none of them has a faceted visualization framework such as we are suggesting. VisFacet makes three main contributions: 1. It adds the visualized, interactive search capabilities of the facets to the VuFind system. The view includes objects filtered by a variety of similar topics and by facets such as Era and Region. 2. It adds a broadened search capability to the visualized view, which is a completely new feature.
3. It contributes the code to the VuFind open source project.
In chapters thirteen and fourteen of their book [
The aim of this research was to enrich the discovery layer of the catalogue with an interactive side-box that presents the facets of the search results graphically. Fig. 1 presents the current textual view as provided by the VuFind package. The term for the search in this example is the word “king.” The area above a set of results presents a list of Suggested Topics that were found in the records addressing the term “king,” such as “history” and “kings and rulers.” The area on the right provides us with the facets—namely, the books taken from the left column filtered by a variety of subjects. For example, within the “Author” category we can see a cluster that contains all the records (202) from the current set of search results that appear to be written by William Shakespeare. Thus, the search system allows the user to search from one search box and then narrow down the results by clicking on the various facets of the results.
The VuFind information retrieval engine, as well as that of most modern library catalogues, uses the MARC records as the core for its search engine index schema over the ILS core. MARC (MAchine-Readable Cataloging) is the de-facto standard for the bibliographical records with metadata for each item (that is, book, journal, movie, and so on). The metadata are inserted as indexes in the VuFind internal database, along with their associated metadata. Given a term to search, the catalogue’s search engine retrieves associated records from the search engine. At the same time, it retrieves a list of predefined facets (sometimes also called clusters), each of which represents a logical conjunction of the current set of the results filtered by several predefined fields. For example, the Topic field is represented in the MARC records with the MARC code 650. If we choose to search for the topic “history,” the results list will display the records where the topic field, represented by 6##, holds the value of a number of records that include the field 6## with the word “history.” Choosing another subject like “second world war” will narrow the search, and all the resulting records that contain “history” but not “second world war” will be excluded from the current results list. A visualized discovery box, which is intuitive, easy to interact with, and which integrates smoothly with web interfaces, is an essential addition to these evolving software systems. However, current commercial companies and academic libraries that develop tools for the customization of the discovery layer do not have this capability. Thus, an additional aim of this research was to extend the complex, modern catalogue software package with visualization capabilities while retaining its modularity and performance quality. Any visualization solution must integrate smoothly with the existing software architecture and be capable of using the retrieved information as any other other facet.
Our tool code was integrated into the VuFind open source system so that the VisFacet box is added to the user interface of the catalogue below the right facets section of the SERP (Fig. 1). The location can be changed to above the facets, should the system administrator wish to configure it that way. Section 3.1 describes the targets we addressed in the graphical design of VisFacet. In section 3.2 we describe VisFacet’s modules and how they were integrated into VuFind.
In the example presented in Fig. 2, the topic node that appears in the center of the star-like graph is “History,” a topic suggested by the system because it appears in the largest number of items that came up in a search for the keyword “king.” The rest of the topic nodes (in green), each representing a cluster of results, are connected to it in descending order in a spiral form according to their grading. Era in orange and Region in blue are two additional facets of our visualization. The color of each facet appears in the legend at the bottom. All the colors (including background) and sizes are configurable and adjustable. To save space, only the first word in a term appears near the nodes in the graph. When a user selects a node (by brushing the mouse over it) the full topic of the current node appears below the graph.
Any click on one of the nodes in the graph will narrow (assuming the toggle at the top is on the Narrow position) the current search results list according to the specific Topic, Era, or Region that was clicked. For example, clicking on the “France” Region facet will drop all records that do not contain France as their region. The Narrow toggle that appears at the top of the box is on by default, while the Discovery toggle is dimmed. Clicking on any of the nodes when Discover is on will trigger a completely new query with the clicked topic as a keyword. Since all of those topics were retrieved from the search engine index, this “Discover” function will never lead to an empty results set. The graph is redrawn with each search operation, whether text or graphic. Readers of this paper are welcome to try it at http://hufind.huji.ac.il/.
The visualization of the suggested topics and the facets should help the end user
discover a subtopic to focus on from a wide-area topic. To address this
requirement we have applied several dimensions to the two-dimensional layout of the
infographic box as follows:
1. Link-Node form: A star-like graph where the topics are connected with an
explicit link to the center that intuitively highlights the relation between the
topics.
2. Spiral form and order: Another dimension of connectivity arises from the
fact that we have arranged all the topics in a spiral form, according to their
frequency. For example, the topic History that occurs 4915 times in the
resulting items appears in the middle of Fig. 2. It is closely connected with the
kings and rulers node that occurs 2905 times in the resulting items. The other
nodes appear in the spiral according to the number of occurrences that can be
seen in the “Suggested Topics” box on the top of Fig. 1. The benefit of using
a spiral form is that it spreads the topics evenly across any stretched box to
optimize space usage as well as implicitly isolate the terms. Moreover, users
gain a sense of the importance of the topics through their closeness to the
center and thus may navigate between them using the spiral route.
To draw the spiral, we redesigned Vogel’s formulation [
Incorporating a visualization tool into the VuFind system, which is a generic engine for building and customizing library catalogues, is a natural evolvement. The VuFind system has complicated software engineering and many code lines. Each part of the system is implemented in a different programming language that is best suited for its particular feature. Hence, the architecture and programming languages of VisFacet were selected and designed in terms of code coherency and performance. The VuFind architecture, shown on the left side of Fig. 3, consists of an application core and two main layers. The data layer contains a search engine index that can be distributed among several machines, or even different libraries, and be updated daily. The application core that runs on the server side and is responsible for bringing data from the data layer performs all the required processing and then passes it to the user interface layer. Finally, the user interface layer is responsible for arranging the data. The VisFacet visualization subsystem is divided into three main components, each incorporated into the appropriate layer of the VuFind system, as can be seen in Fig. 3. The retrieval module, was written in PHP, retrieves the data from the search engine index. It obtains information about a user’s current search, performs its own internal processing, and then returns all the required information needed to render suggestions. This information is processed and organized for the visualization in the Glue UI module that runs in the browser on the client side and was written in Javascript. The Glue UI module binds its own functions to the visualization module and “listens” to interactions. The visualization module gets the organized data and translates it into the visualization objects, which will be presented graphically in a web browser. We incorporated the Processing.js package (http://www.processingjs.org) that is used to create images, animations, and online interactions by using the visual programming language named Processing. It also converts the Processing code into Javascript, thus allowing it to be run by any HTML5-compatible browser, including mobile browsers and current versions of Firefox, Safari, Chrome, Opera, and Internet Explorer. We implemented the visualization module including the spiral algorithm, described in 3.1, in the Processing proprietary language. Behind the scenes, the Processing.js library compiles our code to pure JavaScript.
The VisFacet package was set up, integrated, and tested for two environmental systems: HUfind, which is a customized project based on VuFind, and VuFind itself. After it had passed HUJI library’s approval regarding the usage patterns and performance tests, it was added to the production version of HUfind in May 2014. The site at http://hufind.huji.ac.il/ allows thousands of HUfind users to use a graphic search. Note that it is enabled only when using browsers other than Internet Explorer. To learn more about user satisfaction, we wrote a first-stage evaluation questionnaire, which is described in this section. We designed the questionnaire to measure the experiences of users wishing to discover a new research topic to focus on. The questionnaire was sent to two groups of users: (1) an undergraduate student group composed of 21 students, and (2) a group of librarians composed of 10 experienced librarians who are familiar with the HUfind catalogue. The questionnaire itself consists of two search tasks: the Text search that asks the user to search the suggested topics for the keyword journalists via the regular text interface and the Graphic search that requires the use of our infographic box. Each task requires the exploration of the Suggested Topics and the Region facet in order to focus on a direction. The search experience is questioned in each path with questions that appear in Table 1. The user is asked to compare both experiences and determine whether she prefers one, the other, or a combination of the two.
The columns in Fig. 4 present the average rate of the answers to Q1, Q2, and Q3, according to the group involved: librarians or students. On average, both groups preferred the Text search over the Graphic search. The answers for the graphic evaluation were mainly gave some ideas or very little. The librarians, who are used to text searches, were more skeptical about the graphic visualization, while the students liked it more than the librarians. Some of them preferred the graphic visualization, and therefore the standard deviation for their answers was higher. However, 16 students (76%) answered, in Q6, that they prefer having both types of search. This shows that they are open to this direction but want an improved display. We received the following two comments about VisFacet: (1) the search is less user friendly because it requires brushing long words with the mouse to make them fully visible, and (2) the importance of the order of nodes in the spiral form was not intuitively understood. We will relate to both problems in the next version. The number of steps taken by the student group in both types of search were very similar, 1.91 steps on average, with a standard deviation of 0.80-0.84. The number of steps taken by the librarians was 1.8 steps on average, which is slightly smaller than in the other group, and it was the same for both types of search. Since the questionnaire is not a real discovery task, it is difficult to draw conclusions from this result. To gain greater benefit from VisFacet, we plan to provide additional guidance for its use as well as monitor its use and find out how we can improve it.
We have implemented a subsystem that adds an interactive visualized and faceted search to a modern catalogue. Because our subsystem is a new feature of VuFind, we have explored the existing system to understand its current state and find the appropriate solutions for our design. VisFacet is an initial suggestion for faceted visualization that can be a real impetus for extending the visualization capabilities of modern library catalogues.
We thank Demian Katz, the founder and main developer of the VuFind system, for his ongoing help in enabling the smooth integration of VisFacet into VuFind. We also thank Edith Falk, the chief librarian of the Hebrew University for all her support in this project, and Eli Hayun, the programmer at the Library Authority of HUJI for helping with the integration with HUfind. We thanks Nurit Baltiansky, Mally Cohen, and Avi Allalouf for the discussions regarding the questionnaires.