Information Personal Management - A SIGIR 2006 Workshop
Visualizing a Personal Social Network of Email Archives For Re-Finding
Xiaoyan Yu, Pengbo Liu, Mohammad Alkandari, and Manuel A. Prez-Quiones
Department of Computer Science Virginia Tech Blacksburg, VA 24060
{xiaoyany, pengbo, mkandari, perez}@vt.edu ABSTRACT
Email has become an important communication medium that people routinely rely on to achieve their daily work. However, managing email accounts for life-long communication is not an easy task. Users have different strategies for archiving their messages and have varied understanding and expectations of retrieval and how to use that capability. We have designed and implemented a tool, VisPEAM that applies the benefits of visualization of social network in personal email archive management. This helps users refind email messages by using peoples name, folder and subjects as search criteria. The graphical representation helps users to find their old emails by exploring the social relationships in different folders as well as shows how frequently those people exchange email messages regarding a particular topic. Moreover, the graphic visualization helps people manage their growing collections of email messages with respect to their social relationships. addition, instead of relying on keyword search to get the desirable information, people tend to use small, local steps and context search cues especially, time and other peoples names to re-find information in email [3]. We developed VisPEAM, an add-on feature to email clients. It visualizes social network on personal email archives to uncover the social flavor of the architecture. It also arranges the search results in the context of social network to shorten the steps in building a re-finding chain. The paper is organized as follows. We will first introduce related work, followed by a brief description of personal email archives. We develop scenarios to illustrate the benefits of our solution then focus on how our tool is designed and implemented. Finally we conclude our work and point out future work.
2. RELATED WORK
Many studies focus on issues relevant to email archive management such as email overloading and re-finding. For example, Whittaker et Categories and Subject Descriptors al. [1] proposed technical solutions for the email overload problem, H5.m. Information interfaces and presentation (e.g., HCI): in which they presented qualitative and quantitative information Miscellaneous. about the use of email for task management, personal archiving, and asynchronous communications. The research of Gwizdka [4] seeks to gain understanding of individual differences in email behavior. General Terms This study presented results from a questionnaire-based study that Management, Design, Human Factors discusses how users manage their email messages and how they keep the important messages for future retrieval.
Keywords
Email management, social network, visualization.
1. INTRODUCTION
In societies where the technology is readily available, email has become a universal communication medium in peoples daily life. Their email archives reflect their social lives and also the stories of their happiness and sadness while current email clients hide that social information. The high volume of email messages and unsuccessful filing strategies may hinder people to re-find email from their personal archives. Bush in 1945 envisioned a personal information management tool, MEMEX. It could not only keep all the information but also make good use of them [2]. However, current personal email archive management tools are more successful in supporting email storage than in later re-use. In
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. SIGIR06, August 6-11, 2006, Seattle, Washington, USA. Copyright 2006 ACM 1-58113-000-0/00/0004$5.00.
Implementing visualization in email tools is highly recommended [5] as a way to help users understand the relationships between the various messages as well as the relationships between different people.. For example, Venolia and Neustaedter [6] described thread structure of a person's email and how it is evolved along time.[7] Mansmann et al. [8] applied Self-Organizing Maps (SOM) to visualize archival emails with the feature of important terms in the Subject field. Visualization of social networks has a good history in which many people have been interested in applying social network in email archive management, especially in conjunction with other features such as time. Vigas [9] discussed important concepts in which, both Post History and Social Network Fragments (SNF) focus on two major dimensions of email archive: people and time. Their visualization provides striking insights into the experience of social connectedness over time. These insights basically enable users to better manage how they invest their time and energy into personal relationships as well as improve the whole sense of wellbeing. Our visualization is very like SNF but more focuses on facilitating re-finding email. Some studies also focus on visualization of social network on areas other than email. Vizster [10] is designed and implemented a visualization system for playful end-user exploration and navigation
116
Personal Information Management - A SIGIR 2006 Workshop
of large scale online social networks. Our implementation of information. Each folder contains a number of messages usually VisPEAM is based on Vizster, built on the Prefuse package [12]. relating to the same subject. Therefore, we consider people who appear in the same folder to be connected by the same topic. If there are a great number of common names appearing in two 3. PERSONAL EMAIL ARCHIVE Whittaker [1] observed three different email filing strategies: no- folders, the folders have a high degree of correlation. filers who do not make use of any folders and keep a huge incoming email box; frequent-filers who attempt to file daily and have folder structured archives; and spring-cleaners who occasionally file and usually have an extensive folder structure. People with different filing strategies have different email archive structures. We therefore also take into account, folder information (if it is available) when analyzing relationship in an archive. Our visualization tool is tested on Enrons corpus. The Enron corpus was made public after legal investigation. After removing duplicated messages, there are a total of 200,399 messages belonging to 158 users. [11] We selected a single person Sallys email. In her email archive, all email messages have been organized in predefined folders.
3.2 People Relationship
To extract peoples relationship from email folders, we take advantages of email header information, as all people involved in email threads appear in either the From, the To, or the Cc fields. Between two people, say Person A and Person B, who appear in an email header, there can be five types of relationships: FromTo( A sends email to B or A receives email from B), FromCc( A carbon copies email to B or A receives a copy from B), ToTo( A and B receives the same email ), CcCc( A and B receive the same copy ), ToCc( A receives a email message and B receives a copy or vice versa). These five types of relationships play different roles in depicting peoples closeness. For example, a name which appears in the To field of an email has a stronger relationship with the name in From field than a name in the Cc field.
3.1 Folder Relationship
A natural grouping of people is through the use of folder
Bector McLoughlin
salary
salary
Bector McLoughlin
salary salary
Figure 1. VisPEAM tool: 1) view panel; 2) control panel with 2a) search filter with search for salary and 2b) connectivity filter; 3) result panel shown two matched emails for the selected correspondent BectorMcLouglin.
117
Personal Information Management - A SIGIR 2006 Workshop
supported: search filter and connectivity filter. The result panel Before describing our tool, we will illustrate some scenarios to shows the search results. explain the benefits email archive owners such as Sally can get 5.1 Visual Items from a tool like ours. A personal social network consists of nodes and edges. The Scenario 1: Sally might discover unexpected persons with the help characteristics of the network are summarized in Table 1. of the interactive, visualized social network. When exchanging Correspondents and email folders are presented as nodes. The same emails, a sender might carbon copy to some related people that correspondent in different folders is presented as different nodes. Sally has not realized. The visualized networks such as those shown The attributes of a correspondent node include the persons name in Figure 1 can give Sally an idea of other people involved in the and the closeness (explained in the section 3.2) with the owner. A email. Highlighting neighbors as shown in Figure 2, one of our correspondent node is rendered as a label with this persons name in tools features, would also help Sally, if she were interested in those the text field. The archive owner is not put into the network at all to related people. reduce trivial edges since everyone is connected to the owner. Scenario 2: Sally thinks that she has filed a particular email in the Instead, the text font of a correspondent node is bold or regular to folder book_administrators but does not find it there. By using our present their closeness to the owner. The attributes of a folder node tool, she can explore the neighboring folders to see whether that include the folders name, a folder image to distinguish it with a email is in those or folders not. In addition, she might get to correspondent node. A folder node is rendered as a label with the remember the sender, Brent A. Price, of email during the folder name in the text field and the image in the image field. The exploration. When she searches for Brent, the name would appear book_administrators node in Figure 2 is a folder node. An edge in different places and she then can click on the best fit one (for connecting two correspondent nodes represents the existence of example, the one resides in the aec folder shown in Figure 2 which relationship between the two people. The closer these two people it is more possible in her mind to file the email) to fast locate that are, the thicker the edge is. An edge between two folder nodes email. presents common correspondents in these two folders. The width of such an edge also represents the relative number of the common correspondents residing in the two folders. There are also edges between folder nodes and some of correspondent nodes to visualize correspondents in different folder. For example, all the correspondents in the clique shown at the bottom of Figure 2 have emails in the folder aec.
4. SCENARIOS
Table 1. Summary of visualized items. Items Types Correspondent Node Figure 2. Layout of personal social network with Brent A Price focused on. Folder Closeness Folders name Folder distinguisher Text font Text field Image Attributes Persons name Representation Text field
CorrespondentEmail traffics Edge width Scenario 3: Sally might not remember who has sent a message, say correspondent relevant to a meeting. She roughly recalls the topic. Using this Common Folder-folder Edge width topic, our tool will highlight relevant people shown in Figure 1 in Edge correspondents red. This might help her recall the sender of this message. If there Whose email are too many people highlighted to help recall, she might select CorrespondentA very light filed in which people most likely to be involved in that message. For example, the folder edge folder center of a communication community, such as Bent A. Price in the book_administrators folder who communicates with most people 5.2 Layout in that folder shown in Figure 2, might be a good choice. The network is arranged in the force-directed layout, where each node repels each other and links act as springs. People with more 5. VISPEAM links among them tend to cluster together and naturally form a Our tool, Visualization for Personal Email Archive Management community. People within the same folder are closer to each other (VISPEAM), has as a goal to help an email archive owner explore so it is easily to recognize who are in a particular folder. Each his/her email archive and re-find desired information. We apply community organized by folders has its own pattern. Some of them visualization to a derived social network and integrate a search are cliques such as the aec folder in Figure 2. Some contain functionality. The tool consists of 3 components: view panel, several sub-communities with articulate correspondent nodes such control panel, and result panel, shown in Figure 1. The view panel as the book_administrators folder in Figure 2. These patterns renders a personal social network. The filters on the control panel would help the owner dig out information from the archive, as can highlight and hide some portions of the network. Two filters are illustrated in the third scenario in the section 4.
118
Personal Information Management - A SIGIR 2006 Workshop
5.3 Interaction
A user may be overwhelmed upon facing a complex social network even when depicting his/her own email archive. We hence provide plenty of interactions on the network for users to explore their archives. Our tool supports basic mouse interactions such as panning and zooming. Mouse across a node fills its text field into grey color and changes the text color from black to blue. In this way, as shown in Figure 2, the name is much clearer to recognize than the one without fill-in color. Single left-click on a correspondent node is to select the node for further interaction. Double left-click on a persons name will show his/her email excerpts matching search criteria. Double left-click on an edge between correspondent nodes will show excerpts from the emails exchanged.
original sender was forgotten, or discover new relationships by exploring the social network. The proposed tool has a simple user interface to facilitate the re-finding process. This interactive user interface can highlight the same person in different folders, change focus on the social network, and zoom-in on individual folders. In the future, we plan to conduct a usability study to verify the effectiveness of our tool by observing how users interact with the tool given the above scenarios. We expect to gain further insight into the utility of this tool based on that study.
7. ACKNOWLEDGMENTS
We thank Travis Rose and Uma Murthy for their comments on this paper. We would also like to thank the students in the CS 5984 Personal Information Management course at Virginia Tech for the The owner might be interested in whom a correspondent insightful in-class and forum discussions. communicates with besides himself/herself (see the first scenario in the section 4). The owner can follow each out-link from the person 8. REFERENCES to find all the persons neighbors if it is easy to follow or zoom in [1] Whittaker, S., and Sidner, C. , Email overload: exploring the region around the person. We also support an easier way, personal information management of email. Proceedings of the neighbor highlighter, to help the task. When the mouse is put across SIGCHI conference on Human factors in computing systems: a node, not only the node itself will change but also its neighbors common ground, ACM Press, 1996: p. Pages: 276 - 283 will change from black to green. In Figure 2, it is easy to recognize [2] Bush, V., As we may think, in The Atlantic Monthly. 1945. p. that in the book_administrator folder Brent A. Price communicates 101-108. with everyone but only 2 people Steve Jackson and Scott Mills. [3] Cutrell, E., S.T. Dumais, and J. Teevan, Searching to eliminate personal information management. Commun. ACM, 2006. Our tool supports 3 search filters: name filter, content filter, 49(1): p. 58-64. advanced filter (content plus date). The name filter highlights nodes whose text field prefix match the searching names. The owner [4] Gwizdka, J., Email task management styles: the cleaners and the keepers. CHI '04: CHI '04 extended abstracts on Human would get benefits from the prefix search on text fields without the factors in computing systems, ACM Press., 2004: p. Pages: necessity to complete the names they are looking for. The name 1235 -1238. filter also makes the owner aware of which folders a particular correspondents emails reside. The content filter allows the owner [5] Somervell, J., McCrickard, D., North, C., and Shukla, M., An evaluation of information visualization in attention-limited to search his/her archive by email subject or body and highlights the environments. VISSYM '02: Proceedings of the symposium on nodes of people who are involved in such emails by changing the Data Visualisation 2002., 2002: p. Pages: 211 - 216. color from black to red. Figure 1 shows the results of applying the content filter. The advanced filter can narrow down the search on [6] Venolia, G., and Neustaedter, C., Understanding sequence and reply relationships within email conversations: a mixed-model the arrival date of email messages. Upon applying the content filter visualization. CHI '03: Proceedings of the SIGCHI conference or the advanced filter, excerpts of emails associated with on Human factors in computing systems, ACM Press., 2003: p. highlighted people are shown on the result panel by double clicking Pages: 361 - 368. any of these nodes. An excerpt includes email headers and sentences around highlighted query terms in the email body. [7] Perer, A., B. Shneiderman, and D.W. Oard, Using Rhythms of Relationships to Understand Email Archives. To appear in the Currently we highlight the query terms by adding * around them Journal of the American Society for Information Science and instead of coloring them (see the right panel of Figure 1 for Technology (JASIST), 2006. examples). [8] Mansmann, F., T. Schreck, and D. Keim. MailSOM Our tool runs on Java 2 Standard Edition, version 1.5. The user Exploration of Electronic Mail Archives Using Self-Organizing interface and visualization are built on the Prefuse package [12] Maps. in CEAS. 2005. which offers many useful visualization and interface components. [9] Vigas, F.B., et al. Digital Artifacts for Remembering and After analyzing correspondent and folder relationship from email Storytelling: PostHistory and Social Network Fragments. in archive, a GraphML [13] file describing the social network is Proceedings of the 37th Hawaii International Conference on generated to feed as data to the visualization. The name filter is System Sciences. 2004. implemented by in-memory search. The content and advanced [10] Heer, J., and Boyd, D., Vizster: Visualizing Online Social filters are supported by a search using the Lucene package [14]. Networks. Computer Science Division and School of Information and Systems - University of California, Berkeley., 6. CONCLUSION AND FUTURE WORK 2005. We have developed a tool of visualizing the social network of large [11] Klimt, B. and Y. Yang. Introducing the Enron Corpus in volumes of personal email messages. The benefit of building up a CEAS. 2004. visualized social network lies in its potential ability to reduce the [12] Prefuse, http://prefuse.org/. effort of re-finding email messages. This tool is designed to help [13] GraphML, http://graphml.graphdrawing.org/. people recall necessary information by exploring the social [14] Lucene, http://lucene.opache.org/. network. For example, a user could retrieve messages even if the
119