Tessa A. Lau

Patterns of search: analyzing and modeling Web query refinement

We discuss the construction of probabilistic models centering on temporal patterns of query refinement. Our analyses are derived from a large corpus of Web search queries extracted from server logs recorded by a popular Internet search service. We frame the modeling task in terms of pursuing an understanding of probabilistic relationships among temporal patterns of activity, informational goals, and classes of query refinement. We construct Bayesian networks that predict search behavior, with a focus on the progression of queries over time. We review a methodology for abstracting and tagging user queries. After presenting key statistics on query length, query frequency, and informational goals, we describe user models that capture the dynamics of query refinement.

CoScripter: automating & sharing how-to knowledge in the enterprise

Modern enterprises are replete with numerous online processes. Many must be performed frequently and are tedious, while others are done less frequently yet are complex or hard to remember. We present interviews with knowledge workers that reveal a need for mechanisms to automate the execution of and to share knowledge about these processes. In response, we have developed the CoScripter system (formerly Koala [11]), a collaborative scripting environment for recording, automating, and sharing web-based processes. We have deployed CoScripter within a large corporation for more than 10 months. Through usage log analysis and interviews with users, we show that CoScripter has addressed many user automation and sharing needs, to the extent that more than 50 employees have voluntarily incorporated it into their work practice. We also present ways people have used CoScripter and general issues for tools that support automation and sharing of how-to knowledge.

Programming by Demonstration Using Version Space Algebra

Programming by demonstration enables users to easily personalize their applications, automating repetitive tasks simply by executing a few examples. We formalize programming by demonstration as a machine learning problem: given the changes in the application state that result from the users demonstrated actions, learn the general program that maps from one application state to the next. We present a methodology for learning in this space of complex functions. First we extend version spaces to learn arbitrary functions, not just concepts. Then we introduce the version space algebra, a method for composing simpler version spaces to construct more complex spaces. Finally, we apply our version space algebra to the text-editing domain and describe an implemented system called SMARTedit that learns repetitive text-editing procedures by example. We evaluate our approach by measuring the number of examples required for the system to learn a procedure thatworks on the remainder of examples, and by an informal user study measuring the effort users spend using our system versus performing the task by hand. The results show that SMARTedit is capable of generalizing correctly from as few as one or two examples, and that users generally save a significant amount of effort when completing tasks with SMARTedits help.

End-user programming of mashups with vegemite

Mashups are an increasingly popular way to integrate data from multiple web sites to fit a particular need, but it often requires substantial technical expertise to create them. To lower the barrier for creating mashups, we have extended the CoScripter web automation tool with a spreadsheet-like environment called Vegemite. Our system uses direct-manipulation and programming-by-demonstration tech-niques to automatically populate tables with information collected from various web sites. A particular strength of our approach is its ability to augment a data set with new values computed by a web site, such as determining the driving distance from a particular location to each of the addresses in a data set. An informal user study suggests that Vegemite may enable a wider class of users to address their information needs.

Automatically classifying emails into activities

Email-based activity management systems promise to give users better tools for managing increasing volumes of email, by organizing email according to a users activities. Current activity management systems do not automatically classify incoming messages by the activity to which they belong, instead relying on simple heuristics (such as message threads), or asking the user to manually classify incoming messages as belonging to an activity. This paper presents several algorithms for automatically recognizing emails as part of an ongoing activity. Our baseline methods are the use of message reply-to threads to determine activity membership and a naïve Bayes classifier. Our SimSubset and SimOverlap algorithms compare the people involved in an activity against the recipients of each incoming message. Our SimContent algorithm uses IRR (a variant of latent semantic indexing) to classify emails into activities using similarity based on message contents. An empirical evaluation shows that each of these methods provide a significant improvement to the baseline methods. In addition, we show that a combined approach that votes the predictions of the individual methods performs better than each individual method alone.

DocWizards: a system for authoring follow-me documentation wizards

Traditional documentation for computer-based procedures is difficult to use: readers have trouble navigating long complex instructions, have trouble mapping from the text to display widgets, and waste time performing repetitive procedures. We propose a new class of improved documentation that we call follow-me documentation wizards. Follow-me documentation wizards step a user through a script representation of a procedure by highlighting portions of the text, as well application UI elements. This paper presents algorithms for automatically capturing follow-me documentation wizards by demonstration, through observing experts performing the procedure. We also present our DocWizards implementation on the Eclipse platform. We evaluate our system with an initial user study that showing that most users have a marked preference for this form of guidance over traditional documentation.

Privacy interfaces for information management

To facilitate the sharing of information using mode rn communication networks, users must be able to decide o n a privacy policy—what information to conceal, what to reveal, and to whom. We describe the evolution of pr ivacy interfaces—the user interfaces for specifying privac y policies—in COLLABCLIO, a system for sharing web browsing histories. Our experience has shown us that privac y policies ought to be treated as first-class objects: po licy bjects should have an intensional representation, and priv acy interfaces should support direct manipulation of thes e objects. We also show how these conclusions apply to a varie ty of domains such as file systems, email, and telephony.

Automated email activity management: an unsupervised learning approach

Many structured activities are managed by email. For instance, a consumer purchasing an item from an e-commerce vendor may receive a message confirming the order, a warning of a delay, and then a shipment notification. Existing email clients do not understand this structure, forcing users to manage their activities by sifting through lists of messages. As a first step to developing email applications that provide high-level support for structured activities, we consider the problem of automatically learning an activitys structure. We formalize activities as finite-state automata, where states correspond to the status of the process, and transitions represent messages sent between participants. We propose several unsupervised machine learning algorithms in this context, and evaluate them on a collection of e-commerce email.

Programming by demonstration: an inductive learning formulation

Although Programming by Demonstration (PBD) has the potential to improve the productivity of unsophisticated users, previous PBD systems have used brittle, heuristic, domain-speci c approaches to execution-trace generalization. In this paper we de ne two applicationindependent methods for performing generalization that are based on well-understood machine learning technology. TGenvs uses version-space generalization, and TGenfoil is based on the FOIL inductive logic programming algorithm. We analyze each method both theoretically and empirically, arguing that TGenvs has lower sample complexity, but TGenfoil can learn a much more interesting class of programs.

Sheepdog: learning procedures for technical support

Technical support procedures are typically very complex. Users often have trouble following printed instructions describing how to perform these procedures, and these instructions are difficult for support personnel to author clearly. Our goal is to learn these procedures by demonstration, watching multiple experts performing the same procedure across different operating conditions, and produce an executable procedure that runs interactively on the users desktop. Most previous programming by demonstration systems have focused on simple programs with regular structure, such as loops with fixed-length bodies. In contrast, our system induces complex procedure structure by aligning multiple execution traces covering different paths through the procedure. This paper presents a solution to this alignment problem using Input/Output Hidden Markov Models. We describe the results of a user study that examines how users follow printed directions. We present Sheepdog, an implemented system for capturing, learning, and playing back technical support procedures on the Windows desktop. Finally, we empirically evalute our system using traces gathered from the user study and show that we are able to achieve 73% accuracy on a network configuration task using a procedure trained by non-experts.