The Workflow of Back-end

For converting the review data into useful information, the back-end of Bookle server runs the calculus processing as following workflow:

Step 1 : Extract keywords from each review. Bookle utilizes “Yahoo! Term Extraction tool” (mentioned in 2.1.1) to extract the representative keywords from reviews and store them into server database. If keywords is more than 20, Bookle records the first 20 keywords as Figure 4.2 shown.

Figure 4.2: Keywords Extraction

Step 2-A : Calculate the average value of reading behaviors for each book. To rate review quality, Bookle acquires three different average values from reading behav-iors. The first value is “Average Experience Period” that is a time period in days as the unit. The “Experience Period” is the difference of Date of F inish at and Date of Create at (mentioned in 3.2). The value of “Experience Period” does not mean the actual reading time that user spent for a book. The value could be the best indicator for measuring how users love the book. When readers are very inter-ested in a book, they would desire to go through whole book as soon as possible.

The second value is the amount of “Average Bookmarks” of the book. Users could

set a bookmark at specific page in E-book for recording reading progress or just making a mark. The amount of bookmarks is also one of criteria for rating review quality. The last one is the amount of “Average Notes” in a book. Users could make the section that they feel interested in with a circle or an underline in pages.

The amount of notes represents how seriously users read the book, and it is also a good reference for rating review quality.

ExperienceP eriod(EP ) = Date of F insih at− Date of Create at (4.1) If there are n reviews for a book, let R_i be the review i. Let EP_i, BM_i, and N ote_i represent the Experience Period, Bookmark number, and Note number respectively.

The formula for calculating the average value of reading behaviors can be expressed as following:

Step 2-B : Measure the quality for each review. The main assessment method bases on the data about reading behaviors. There are four criteria for Bookle to measure the review quality.

1. Base Quality(BQ): The “Base Quality” is measured by the value about “Ex-perience Period”. This is the most obvious indicator to determine the degree how readers love one book, and this value is regarded as the most influential variables of review quality. When a reader likes a book very much, the value of Experience Period will be very small, even though he or she is in a busy life. If the Experience Period that someone spent on a book is ten times faster than

calculates the Base Quality by the ratio of average Experience Period and per-sonal Experience Period. The Base Quality value of a review is calculated as Formula 4.5 shown.

BaseQuality(BQ) = avg.EP

P ersonalEP (4.5)

2. Bookmark Quality(BmQ): The value of “Bookmark Quality” is derived from the amount of bookmarks. In the process of how to decide the value of Book-mark Quality, Bookle consider the relationship about the Experience Period and bookmark numbers. For example, Kevin has a lot of leisure time to do things in which he interested. Hence, he could read his favorite books in a short period of time. From his reading behaviors, the “Experience Period” is seven times faster than the average and he made few bookmarks. For another example, Alice is a busy office worker, and she must make good use of spare time on her interests. Hence, it took her longer time period than Kevin to read through the same book. From her reading behaviors, the “Experience Period”

is three times faster than the average and she made many bookmarks. If Bookle only consider the value of “Experience Period”, the fronter has better review quality. However, Alice made more efforts on reading the book than Kevin, so the later should have better review quality. It should be realized that Bookle need to provide the additional quality score from the data of bookmarks for the later. Generally speaking, a review with more bookmarks is valued with higher Bookmark Quality in the case of the same Experience Period. To obtain the Bookmark Quality, Bookle calculates the quotient of personal bookmark num-bers and average bookmark numnum-bers for relative comparison. The complete value of “Bookmark Quality” is the calculation about “Experience Period” and

“Bookmark Number” as Formula 4.6 shown.

BookmarkQuality(BmQ) = avg.EP

P ersonalEP ∗ P ersonalBM

avg.BM (4.6)

3. Note Quality(NQ): The expression of Note is a circle or an underline on pages for making a key block. The readers who make notes while reading are usually more diligent, but personal reading habits and types of books are also factors which should be taken into consideration. It generates more positive effects

Bookle measures Note Quality by comparing personal note numbers with the average note numbers of a book. If personal note numbers are ten times more than the average value, it would be scored full marks in this criteria.

N oteQuality(N Q) = P ersonalN ote

avg.N ote (4.7)

4. Favorite Quality(FQ): If readers put a book into virtual bookcase “My Fa-vorite”, the review quality would be rewarded extra bonus.

In conclusion, the total value of review quality is the sum of the above four criteria as Formula 4.8 shown. We can adjust the constant terms of the four criteria dy-namically by situation. The complete measurement of review quality is expressed as Figure 4.3 shown.

ReviewQuality = αBQ + βBmQ + γN Q + δF Q (4.8)

Figure 4.3: Review Quality Measurement

Step 3 : Calculate keyword weight for each keyword in each book. Keywords are ex-tracted from reviews of books at Step 1, and Bookle would assign a weight for every keyword as Figure 4.4 shown. Bookle calculates the weight of each keyword by occurrences and quality of reviews which the keywords occurred. Same keywords keywords might reoccur in different reviews of one book, and the quality values of different reviews combine into the weight of the keyword. In generally, the more

weight of a keyword by the sum of the review quality value from which the keyword is extracted. For keyword T of the book B, if keyword T reoccurs in n reviews R_n and the review quality of Rn is Qn, the weight of keyword T for this book B is the sum of Q_n as Formula 4.9.

W eight(T ) =

∑n i=1

Q_n (4.9)

Figure 4.4: The Weight for Keywords

Step 4 : For each keyword, rank the books that related to the keyword. At step 3, Bookle obtains information about keyword weights of all books. At this step, Bookle utilizes the information to rank books for keywords. The same keywords might have relation with different books, because readers have the same experience between the books.

The same keywords in different books would be weighted in different value. Bookle ranks books for each keyword by sorting the top-K weights. After this step, Bookle obtains information about the top-K books for each keyword as Figure 4.5 shown.

Figure 4.5: The Books Related to Keywords

Finally, Bookle can make the suggestion by the information from Step 4 when users request for recommender service. The information about books related to keywords could help Bookle to speculate what books users want. This step is also the final step of back-end