PowerBookmarks: A System for Personalizable Web Information Organization, Sharing, and Management

Wen-Syan Li, Quoc Vu, Edward Chang, Divyakant Agrawal, Kyoji Hirata
Sougata Mukherjea Yi-Leh Wu, Corey Bufi, Kevin Chang, Yoshinori Hara
C&C Research Laboratories, NEC USA, Inc.
110 Rio Robles, M/S SJ100, San Jose, CA 95134, USA
Phone: (408)943-3008 Fax: (408)943-3099)

Reiko Ito, Yutaka Kimura, Kazuyuki Shimazu, Yukiyoshi Saito
5th Development Department, 2nd Computers Software Division, NEC Corporation
1-10, Nisshin-Cho, Fuchu-Shi, Tokyo 183-8501, JAPAN

1. Introduction

Users of the Web are overloaded with information. This medium is ``polluted" with redundant, erroneous and low quality information. It has no aggregate structure which organizes distinct Web localities. Users have no global view of the entire Web from which to forage for relevant pages. As a matter of fact, a WWW survey of 11,700 users conducted from April 10 to May 10, 1996[1] indicates that 30.31% of the users report "finding known info" is a problem and 27.80% and 12.16% of the users report organizing collected information and finding pages already visited as their significant problems respectively.

**Figure 1:** WebDB Functionality Overview
$\begin{figure}\centerline{\psfig{figure=overview1.eps,width=186mm}}\vspace{-5pt}\end{figure}$

Another study[2] on user behavior of using bookmarks indicates that users gradually build a small sized archive where 68% of users have 11 to 100 bookmarks Over 93% of users create 0 to 5 bookmarks in each browsing session. It also found that a larger archive requires more sophisticated means of organizing. However, existing Internet browsers only provide a flat layout format for bookmarks, which has low usability and poor scalability. An empirical study[3] on users' revisitation patterns to WWW pages found that 58% of an individual's pages are revisits. With these study results, we believe the Web users would like to build and organize a larger collection of bookmarks for future references than they can reasonably maintain now.

PowerBookmarks is being developed under the auspices of the WebDB project[4,5] to address the issues above. Unlike existing bookmarking schemes, where bookmarks cannot be shared, PowerBookmarks supports information sharing with access control. PowerBookmarks also support automated bookmark classification based on document contents as well as many useful personalizable services, such as automated dead link and inactive link removal and new or modified document subscription services. We start with the overview of WebDB, followed by the functionalities of PowerBookmarks with demonstrative window dumps.

2. WebDB Hypermedia Database System

The Web is a collection of multimedia documents in the form of HTML pages connected through hyperlinks. Unlike most search engines, which focus on information retrieval based on keywords, WebDB aims at supporting database-like comprehensive query functionalities as well as navigation, and document generation functionalities with customizability. It is designed to support enterprise level document management, Web site page information maintenance, and to support hypermedia database functionalities, we augment the traditional concepts of tables in relational databases and classes in object-oriented databases with notions of document formats and navigation. We design WQL (Web Query Language)[5] as an HTML document manipulation language. WQL language statements contain two parts: SELECT...FROM...WHERE clauses for specifying retrieval of data contents from hypermedia databases and CREATE...AS... clauses for specifying the output HTML format and navigation of the query results.

The WebDB hypermedia database system provides storage, object-oriented document modeling, SQL3 like document language query processing, and HTML/VRML document generation for PowerBookmarks. The architecture of WebDB has two levels: a logical level consisting modules for logical Web modeling, the WQL query language, and HTML/VRML document generator; and a physical level consisting modules for internal class representations, query processing, and the query result class generator. Figure 1 gives an overview of WebDB architecture and its design is highlighted as follows:

HTML documents are logically modeled as object-oriented hierarchical structures while being physically modeled and stored in the underlying NEC PERCIO OODBMS as classes. A logical/physical schema mapping is maintained for query translation.

**Figure 2:** Internet Query Window
$\begin{figure}\centerline{\psfig{figure=inter_q.ps,width=2.in}}\vspace{-10pt}\vspace{10pt}\end{figure}$

**Figure 3:** Internet Query Results
$\begin{figure}\centerline{\psfig{figure=inter_query.ps,width=3.5in}}\vspace{-10pt}\vspace{10pt}\end{figure}$

WQL is used for querying WebDB. A visual query interface is supported to assist users in specifying queries while actual WQL queries are generated automatically by the WQL query generator. Hence, the complexity of the underlying schema and the query language remains transparent to the user.

**Figure 4:** Query Interface
$\begin{figure}\centerline{\psfig{figure=query.ps,width=3.5in}}\vspace{-10pt}\vspace{10pt}\end{figure}$

**Figure 5:** Browsing Detail Document Metadata
$\begin{figure}\centerline{\psfig{figure=detail.ps,width=3.5in}}\vspace{-10pt}\vspace{10pt}\end{figure}$

**Figure 6:** Navigation through keyword anchor *Database*
$\begin{figure}\centerline{\psfig{figure=data_kw.ps,width=3.5in}}\vspace{-10pt}\vspace{10pt}\end{figure}$

The SELECT..FROM..WHERE clauses in WQL are then translated into their corresponding internal query tree representations for processing against the object-oriented class schema. The translation is performed by the WQL parser by consulting the logical/physical schema mapping.
After query processing, the query results returned by the NEC PERCIO OODBMS are in the internal object-oriented class format. The HTML/VRML document generator of WebDB then converts the query results in internal representations to their corresponding HTML/VRML forms.
Various forms of query result representations, such as HTML tables and VRML visualization, are supported. Anchors in the HTML result documents are implemented as queries to WebDB. Similarly, each node in the VRML visualization page is implemented as an anchor with embedded queries to WebDB to generate new VRML pages as navigation destinations.

**Figure 7:** Bookmark Structure
$\begin{figure}\centerline{\psfig{figure=stru.ps,width=3.5in}}\vspace{-10pt}\vspace{10pt}\end{figure}$

**Figure 8:** Navigation through Classification Hierarchy (Root $\rightarrow$ Computer $\rightarrow$ Software)
$\begin{figure}\centerline{\psfig{figure=software.ps,width=3.5in}}\vspace{-10pt}\vspace{10pt}\end{figure}$

**Figure 9:** Subscription Folder Browsing window
$\begin{figure}\centerline{\psfig{figure=sub.ps,width=3.5in}}\vspace{-10pt}\vspace{10pt}\end{figure}$

**Figure 10:** User Preference Setup Window
$\begin{figure}\centerline{\psfig{figure=prefer.ps,width=3.5in}}\vspace{-10pt}\vspace{10pt}\end{figure}$

3. PowerBookmarks Functionalities

In this section, we summarize some major functionalities of PowerBookmarks as follows:

Querying:PowerBookmarks supports a simplified query interface, shown in Figure 2, for users to query the Internet. After the results are returned by Internet search engines, a filter program extracts useful information and present it to the user. Figure 3 shows the query results for keywords auto and Ford. The user can select a set of URLs to browse in a slide-show fashion or save them as bookmarks in PowerBookmarks. When the user requests to collect URLs into PowerBookmarks, PowerBookmarks performs a sequence of tasks as follows: (1) downloading the URLs; (2) extracting keywords, summary, links, and related URL metadata, such as last modified date, etc.; (3) indexing the URLs, including indexing in a full text search engine TOPIC[6], cooccurring word analysis, and storing metadata in WebDB. After the documents are indexed in PowerBookmarks (through WebDB), the user can issue more complex queries to retrieve bookmark URLs. Figure 4 shows the query interface used in PowerBookmarks. In this window, the user issues a query retrieving the bookmarks related to call for papers for conferences related to XML. Queries with more complex criteria, such as links, full text search, relaxed keyword search[4], are also supported. The query functionalities in PowerBookmarks is carried out by the underlying WebDB.

Browsing and Navigation: The user can also click on a bookmark to view the HTML document or its detail descriptive meta information. Figure 5 shows a window dump for a bookmark URL. Note that at the bottom of this window, there are three comments by the owners who have this URL as a shared bookmark. This functionality allows collaborative work and social filtering in a workplace. In addition to browsing, the user can also navigate through the hypermedia database system of bookmarked HTML documents. In Figure 5, the detail information for a bookmark includes a summary and its most significant keywords. The user can click on the keyword anchor to navigate to other bookmark pages with such a keyword. Figure 6 shows the navigation result when the user clicking the database keyword anchor in Figure 5.

Classification of Bookmarks:PowerBookmarks supports automated document classification by contents based external classifiers similar to Pharos[7]. Figure 7 shows a classification tree for bookmark URLs in PowerBookmarks. Note that this window also displays the number of bookmarks in each folder and the access frequency of each folder (i.e. category) indicated with a temperature icon. The user may navigate through the path to locate URLs of his/her interests. Figure 8 shows a window dump for display of documents and folders at the node Computer/Software. At this node, there are the user's own bookmarks (shared and non-shared) and shared bookmarks owned by other users. The bookmarks with a red cross are dead links detected by the system. The system can also automatically, if the user prefers, move the dead links and inactive bookmarks to the Deleted folder at the bottom of the window. The user can click on any folder to go to other nodes in the bookmark structure.

Subscription Service:PowerBookmarks supports subscription services for new or modified documents on the Internet and intranet as described in [8]. The user can set the subscription criteria, such as temporal, domains, keyword similarity, or document similarity, etc. When there are incoming bookmarks available, the user will be notified (i.e. indicated with a spark) as shown in Figure 9

Preference Setup Window: One unique feature of PowerBookmarks is personalizability. Figure 10 shows that the user sets up his preference, such as automated removal of dead links and inactive bookmarks as well as the depth and the fanout of bookmark classification tree for navigation. The parameter of the period for considering user access pattern is used to adjust the ranking formula for bookmarks in the classification structure and user profile analysis for subscription.

4. Conclusion

This demonstration proposal has highlighted many features of an advanced and personalizable Web information management system. It shows a real and useful application of integrating database and Web technologies. The system demonstration will serve as a forum for stimulating discussion and productive interaction among researchers and users in the area of Web-based information and database system attending 1999 ACM SIGMOD/PODS Conference.

Bibliography

1: Georgia Tech Research Corporation.
2: David Abrams and Ron Baecker.
3: Linda Tauscher and Saul Greenberg.
4: Wen-Syan Li and Junho Shim.
5: Wen-Syan Li, Yi-Leh Wu, Junho Shim, Kyoji Hirata, Sougata Mukherjea, Divyakant Agrawal, Yoshinori Hara, Reiko Ito, Yutaka Kimura, Kazuyuki Shimazu, and Yukiyoshi Saito.
6: NEC Corporation.
7: R. Dolin, D. Agrawal, A. El Abbadi, and J. Pearlman.
8: Wen-Syan Li, Yi-Leh Wu, Corey Bufi, Chen-Chuan Kevin Chang, Divyakant Agrawal, and Yoshinori Hara.