WWW.DISSERTATION.XLIBX.INFO FREE ELECTRONIC LIBRARY - Dissertations, online materials

<< HOME
CONTACTS

Pages:     | 1 |   ...   | 10 | 11 || 13 | 14 |   ...   | 19 |

# «Learning Implicit User Interest Hierarchy for Web Personalization by Hyoung-rae Kim A dissertation submitted to Florida Institute of Technology in ...»

-- [ Page 12 ] --

where time t is the active window duration, the time interval, ti-ti-1, is 100 milliseconds, P(ti) is a mouse location with x and y coordinates at time ti, and the Dist function is a Euclidean distance.

6.1.5. Number of Mouse Clicks People use “click” to hyperlink to another web page. In addition, clicking can be considered as a habitual behaviour (Jung, 2001). Clicking can be a way of expressing our emotions such as if some people are happy to find a product that they were looking for (e.g., book), then they can click the object several times repeatedly. This indicator was examined in Kixbrowser (Jung, 2001), Curious browser (Claypool et al., 2001), Goeck’s browser (Goecks et al., 2000), and Letizia (Liberman, 1995). We use the hypothesis that the greater the number of mouse clicks on a web page is, the more a user is interested in it (Jung, 2001). The number of mouse clicks is counted every time a mouse button is clicked.

6.1.6. Distance of Scrollbar Movement A user can also scroll a web page up and down by dragging a scrollbar. Those dragging events can occur several times while a user is reading a web page. The distance of scrollbar movement for an occasion, E, can be calculated by measuring the mouse movement every 100 milliseconds. By summing all distances of scrollbar movement for all occasions, the distance of a scrollbar movement for a web page can be calculated. The

–  –  –

where E is the number of times the scrollbar is pressed, time E(j) is the duration that the scrollbar is dragged in a single dragging event, and ti-ti-1, is 100 milliseconds. We hypothesize that greater scrollbar movement is correlated with more user interest in a web page.

6.1.7. Number of Scrollbar Clicks The length of many web pages is longer than the height of a monitor. If a user finds a web page interesting, he or she may read further down the web page. A user can scroll down a web page either by clicking or by dragging the scrollbar. Those events are counted separately. The number of scrollbar clicks is counted every time a user clicks scrollbar. As a user scrolls a web page up and down by clicking, the number of scrollbar clicks increases. Jung (2001), Goecks et al. (2000), and Claypool et al. (2001) measured this event and reported that it is a good indicator. We hypothesize that we will also find that the number of scrollbar clicks is correlated with a user’s interest in the web page.

6.1.8. Number of Key UP and Down When scrolling a web page, some people use the “up” and “down” keys instead of the scrollbar. This indicator is similar to the number of scrollbar clicks and the distance of scrollbar movement. The hypothesis is that the greater the number of key up and down presses, the more a user is interested in the web page. This event is measured by increasing the count every time a user strikes up or down keys. Curious browser (Claypool et al.,

2001) and Jung (2001) measured keyboard activities. But they did not measure the key up and down for measuring scrollbar movement.

6.1.9. Size of Highlighting Text While reading a web page, if a user copies some contents of the web page it probably means that the user is interested in the web page. Furthermore, a user can also habitually highlight portions of the page that they are interested in, which is a sign that the user is interested in the page. We assume that the more a user highlights in a web page, the more a user is interested in that web page. A user can highlight several different sentences in a web page for several different occasions. We sum all highlighted contents at the end.

Jung (2001) examined this indicator. He used the Euclidean distance between two points of pressing and releasing. The weakness of his measure resides in neglecting the texts highlighted horizontally when the mouse moves vertically. In order to solve this problem, we assumed a character is 5 pixels, each line has 80 characters, and distance between two lines is 20 pixels on average. The formula is

–  –  –

between two points, and DistX is the horizontal distance between two points.

6.1.10. Other Indicators We also measure other less-frequently-used events such as bookmark, save, print, and memo. A user usually bookmarks web pages in order to visit them later again. We assume those bookmarked web pages are interesting to a user (Li et al., 1999; Maarek and Ben-Shaul, 1996). This can be measured by detecting bookmarking activities during the experiment. Users save important/interesting web pages in their hard drive by using the “Save As” command. This also implies that those saved web pages are interesting to users (Liberman, 1995). This indicator is also counted by detecting saving activities during the users’ browsing. Most web browsers allow users to print web pages. These printed web pages are likely to be interesting to users (Kim et al., 2001). The Memo box is a new feature added in our system. It allows a user to write down a short description on a web page. When the user visits the web page again, the message shows up on the Memo box automatically. We assume that if a user is interested in a web page, then s/he will write a note about the web page.

6.2. Detecting Face Orientation The look at it duration is the time when a user looks at a computer screen. In order to count the time we monitor a user’s head orientation. In this chapter, we detail how we detect the head orientation using a webcam. For detecting head orientation, we use three dots on the hat that a user wears during our experiment. Discussed are how to detect the three dots and how to learn head orientation with the three dots.

6.2.1. Detecting Three Dots To recognize head orientation in an image, the background is usually removed (Intel Inc., 2001). The term “background” stands for a set of motionless image pixels, that is, pixels that do not belong to a human object in front of the camera. The background image can reduce the performance or increase the time complexity.

A simple model of removing background may assume that every background pixel brightness varies independently, according to normal distribution. The background characteristics can be calculated by the mean and standard deviation for every pixel of several dozens of frames collected. After that the pixel in a certain pixel location in certain frame is regarded as belonging to a moving object if condition Abs(mean(x,y)– p(x,y))3×StandardDeviation(x,y) is met, where p(x,y) is a pixel in a new frame (Intel Inc., 2001). However, as the object moves closer to the camera (from “Far” to “Close” in Figure 31), the background color changes from dark color to almost white color -- the assumption is not right (do not follow normal distribution).

Since a user is wearing a black hat, the derivation between two adjacent points helps detect the boundary of the hat. If we continue to remove the pixels that have lower derivation than a threshold starting from edges (left, right, and top) and stop when the derivation is higher than the threshold, at the end the object alone remains without background. The formula for the derivation is

–  –  –

where p(x,y) is the pixel at the coordination of x and y and 200 of the constant is chosen as the threshold by observation. The distance between two pixels is the difference of green color values of two pixels.

Once the background is removed, the image is converted into a black and white image – entire image becomes black except three dots. Each dot/circle is detected by connecting adjacent white pixels as shown in Figure 32. Some erroneous white pixels are cleaned by applying some rules such as removing pixels which size is smaller than 5 and choosing top three circles. The 3 dots are depicted in Figure 33: left, center, and right dots.

a) Far

–  –  –

Figure 33. Three dots in an image 6.

2.2. Learning Face Orientation 6.2.2.1. Input/Output Parameters The information about the 3 dots becomes the input parameter of a learning algorithm (this will be explained later). All examined information about the three dots is {the coordination of left, center, right dots; the sizes of left, center, right dots;

distances between dots; the coordination of center dot; the ratios between the sizes of dots; the ratios between the distances of dots; the angles of left, center, right dots; the levelness of the line between left and right dots}.

The size of a dot is the number of pixels in a dot. If the size of a left dot is smaller and the size of the right dot, then it indicates that the face is not oriented to the web browser. The ratio between the sizes of the left and the right dots can be a more accurate indicator because the ratio is independent from the distance between a face and the monitor. The distances between dots can be used to measure the face orientation. Short distance between the left and the right dots shows that the face does not direct to the monitor. The relative ratios among the three distances are independent from the distance between a face and the monitor. We also use the three angles of three dots and the levelness between the left and the right dots. The levelness will tell us whether the head is up strait or not. All input

values are normalized (between 0.1 – 0.9). The output parameter is the head orientation:

looking at the monitor (0.1) and not looking at the monitor (0.9).

6.2.2.2. Learning Algorithm Artificial Neural Network (ANN) algorithm provides a robust approach to approximating various types of examples. The BACKPROPAGATION algorithm is the most commonly used ANN learning technique. The BACKPROPAGATION algorithm has proven surprisingly successful in many practical problems such as learning to recognize handwritten characters (LeCun et al., 1989), learning to recognize spoken words (Lang et al., 1990), and learning to recognize faces (Cottrell, 1990). This algorithm is appropriate for problems with the following characteristics (Mitchell, 1997): the input and the output should be able to be represented as vectors; the training examples may contain errors; long training times are acceptable; fast evaluation of the learned target function may be required; the ability of humans to understand the learned target function is not important.

Our input and output data can be represented as vectors; error can reside in our training data; training time is not important in our work; since our web browser has to evaluate an image every other seconds, fast evaluation of the learned target function is required; the ability of humans to understand the learned target function is not important.

These characteristics of our example data match BACKPROPAGATION algorithm.

We used BACKPROPAGATION to learn the face orientation with 100 hidden units and

0.03 of learning rate. The results reached more than 90% accuracy. Since the purpose of this paper is not to evaluate the performance of our face-orientation-detection method, we omit the technical details and the analysis of the results.

6.3. Experiments 6.3.1. Experimental Data and Procedures For our experiments, we built a web browser that can record the indicators described above from user’s behaviour and used a camera to record images for identifying face orientation. 11 data sets were collected from 11 different users. Of the 11 human subjects, 4 were undergraduate students, 6 were graduate students, and 1 was a Ph.D.

student. In terms of major, 7 were Computer Sciences, 2 were Aeronautical Sciences, 1 was Chemical Engineering, and 1 was Marine Biology. Each subject was asked to spend a total of 2 hours at the computer. Volunteers were allowed to leave the computer and do other non-computer work. All volunteers were encouraged to behave as normal as possible.

To get a variety of behaviours, we asked the volunteers to divide their activities into multiple sessions, each of which does not exceed 1 hour.

In the browser used in our experiment, most of the functions in Microsoft Explore

6.0 were implemented. The popup windows were disabled initially, but our browser allowed a user to change the option to able them. We asked users to bookmark more than 10 pages, save more than 5 pages, print more than 5 pages, use Memo on more than 5 pages. The browser had Memo box so that users can write small note on a web page. Our web browser takes a picture of a user every 2 seconds. Every time a user leaved a web page, the web browser asked the user how much they are interested in the web page – there were 5 scales between “not interested” (1) and “very interested” (5).

The interests were subjective to each user. The system had a “rescore” button to allow changing the score marked in the previous visit. The browser was written in Visual Studio.NET and ran on a Pentium 4 CPU. The Operating System was Windows XP.

6.3.2. Evaluation Criteria Two evaluation criteria are used: how accurate an indicator could predict a user’s interest and how many users an indicator can accurately predict their interests. Instead of mixing all users’ data sets together, each individual data set was analysed separately so that we could clearly observe whether some indicator predicted certain individual’s interests more accurately than other indicators. An indicator that could predict the score with a lower variance is a more accurate indicator. In order to evaluate each indicator to see which one is more predictable, we use ANOVA (Analysis of Variance). Jung (2001) treated the scale as numeric scale and applied linear regression, multiple linear regression, etc.

methods. We, however, consider the interest scores as discrete values and check if the indicator values are significantly different among the five different interest scores provided by the user. For ANOVA, we use a confidence level of 95% to indicate statistical significance. If the difference is significant, indicator values can predict interest scores. As a second criterion, we count the number of users predicted accurately by an indicator. This criterion indicates how reliable the indicator is across different users.

Pages:     | 1 |   ...   | 10 | 11 || 13 | 14 |   ...   | 19 |

Similar works:

«Does Automotive Service Excellence (ASE) Certification Enhance Job Performance of Automotive Service Technicians? by Emmanuel Kolo Dissertation submitted to the faculty of the Virginia Polytechnic Institute and State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Career and Technical Education Department of Teaching and Learning Approved by: Dr. Curtis Finch, Chair Dr. Gary Skaggs Dr. John Burton Dr. Susan Asselin March 2006 Blacksburg, Virginia...»

«UNIVERSITY OF CALIFORNIA, SAN DIEGO Gravity changes associated with underground injection of CO2 at the Sleipner storage reservoir in the North Sea, and other marine geodetic studies. A dissertation submitted in partial satisfaction of the requirements for the degree Doctor of Philosophy in Earth Sciences By Scott L. Nooner Committee in charge: Mark A. Zumberge, Chair Donna K. Blackman Ahmed-Waeil Elgamal Robert L. Parker Glenn S. Sasagawa Fred N. Spiess 2005 Copyright Scott L. Nooner, 2005 All...»

«Adult Children’s Experiences of Caring for Elderly Chinese Immigrant Parents Managing Chronic Illness: A Qualitative Study by Raza Mehdi Mirza A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Leslie Dan Faculty of Pharmacy Graduate Department of Pharmaceutical Sciences University of Toronto © Copyright by Raza Mehdi Mirza 2013 Adult Children’s Experiences of Caring for Elderly Chinese Immigrant Parents Managing Chronic Illness: A Qualitative...»

«NATION, NOSTALGIA AND MASCULINITY: CLINTON/SPIELBERG/HANKS by Molly Diane Brown B.A. English, University of Oregon, 1995 M.A. English, Portland State University, 1998 Submitted to the Graduate Faculty of Arts and Sciences in partial fulfillment of the requirements for the degree of Doctor of Philosophy University of Pittsburgh 2009 UNIVERSITY OF PITTSBURGH ARTS AND SCIENCES DEPARTMENT OF ENGLISH AND FILM STUDIES This dissertation was presented by Molly Diane Brown It was defended on May 14,...»

«Ph.D. Ordinances and Regulations (Revised up to 84th Meetings of the Senate) ORDINANCES 0.1 The Indian Institute of Technology Guwahati awards the degree of Doctor of Philosophy (PhD) in Engineering/Design/Sciences/Humanities and Social Sciences (HSS) to a candidate who has successfully completed the stipulated Programme of Research. The Programme of Research with the governing rules and regulations are formulated by the 0.2 Senate of the Institute. The Senate can modify or change the...»

«INFORMATION RETRIEVAL VIA UNIVERSAL SOURCE CODING A Thesis Presented to The Academic Faculty by Soo Hyun Bae In Partial Fulﬁllment of the Requirements for the Degree Doctor of Philosophy in the School of Electrical and Computer Engineering Georgia Institute of Technology November 2008 INFORMATION RETRIEVAL VIA UNIVERSAL SOURCE CODING Approved by: Dr. Biing-Hwang Juang, Advisor Dr. Linda M. Wills School of Electrical and Computer School of Electrical and Computer Engineering Engineering...»

«University of Alberta A Magnetotelluric Investigation of the Arabia – Eurasia Collision in Eastern Anatolia by Er an Türko lu A thesis submitted to the Faculty of Graduate Studies and Research in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Geophysics Department of Physics Edmonton, Alberta Spring 2009 ABSTRACT Eastern Anatolia is the location of a young continent – continent collision between the Arabian and Eurasian plates in Eastern Turkey. The first...»

«THE UNIVERSITY OF CHICAGO SOUTHERll BACKWOODS DICTION 18t9-1840 DISSERTATION A SUBMITTED TO THE GRADUATE FACULTY IU CANDIDACY FOR THE DEGREE OF DOCTOR OF PHILOSOPHY DEPARTMENT OF ENGLISH LANGUAGE LI TERATURE AND BY MITFORD McLEOD MATHEWS ILLIUOIS CHICAGO, JUNE, Introduction. From the time of John Witherspoon (1723-1794) there has been in this country more than a little interest in American English as distinguished from the English of England. In 1781 when Witherspoon wrote his articles dealing...»

«THE SAVAGE SELF THE SAVAGE SELF: INDIANS AND THE EMERGENCE OF THE MODERN BRITISH SUBJECT By ROBBIE RICHARDSON, B.A. (HONOURS), M.A. A Thesis Submitted to the School of Graduate Studies in Partial Fulfilment of the Requirements for the Degree Doctorate of Philosophy McMaster University © Copyright by Robbie Richardson, December 2010 DOCTORATE OF PHILOSOPHY McMaster University (English) Hamilton, Ontario TITLE: The Savage Self: Indians and the Emergence of the Modem British Subject AUTHOR:...»

«Colón 1   Mysterious Melodies: Futurism, Primitivism, and the Onomatopoetics of ultraísmo and negrismo David Colón Texas Christian University In his influential essay “Posibilidades creacionistas,” published in Cervantes in October of 1919, Gerardo Diego schematizes the literary concept of the imagen, the most vital compositional facet of the Spanish and Latin American poetic avant-gardes of the 1920s. Diego begins by identifying the various forms of the imagen, rhetorically explained...»

«Cultural Heritage and Contemporary Change Series IVA, Eastern and Central Europe, Volume 39 General Editor George F. McLean Knowledge and Belief in the Dialogue of Cultures Edited by Marietta Stepanyants Institute of Philosophy, Russian Academy of Scienses The Council for Research in Values and Philosophy Copyright © 2010 by The Council for Research in Values and Philosophy Box 261 Cardinal Station Washington, D.C. 20064 All rights reserved Printed in the United States of America Library of...»

«The Relationship Between Phenomenality and Intentionality Comments On Siewert's The Significance Of Consciousness Brie Gertler Department of Philosophy University of Wisconsin 5815 H.C. White Bldg 600 North Park St. Madison, WI 53706 U.S.A. bgertler@facstaff.wisc.edu Copyright (c) Brie Gertler 2001 PSYCHE, 7(17), October 2001 http://psyche.cs.monash.edu.au/v7/psyche-7-17-gertler.html KEYWORDS: Phenomenal character, intentional content, representation, consciousness, what it's like. COMMENTARY...»

<<  HOME   |    CONTACTS
2016 www.dissertation.xlibx.info - Dissertations, online materials

Materials of this site are available for review, all rights belong to their respective owners.
If you do not agree with the fact that your material is placed on this site, please, email us, we will within 1-2 business days delete him.