{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

3 where count gives the number of times the feature

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: ount" gives the number of times the feature appeared in the description of the rated book. Other content-based approaches can present similar explanations 5. After reviewing the recommendations and perhaps disrecommendations, the user may assign their own rating to examples they believe to be incorrectly ranked and retrain the system to produce improved recommendations. As with relevance feedback in information retrieval 28 , this cycle can be repeated several times in order to produce the best results. Also, as new examples are provided, the system can track any change in a user's preferences and alter its recommendations based on the additional information. 3 EXPERIMENTAL RESULTS 3.1 Methodology 3.1.1 Data Collection Several data sets were assembled to evaluate Libra. The rst two were based on the rst 3,061 adequate-information titles books with at least one abstract, review, or customer comment returned for the subject search literature c- Data Lit1 Lit2 Myst Sci SF Number Exs Avg. Rating  Positive r 5 936 4.19 36.3 935 4.53 41.2 500 7.00 74.4 500 4.15 31.2 500 3.83 20.0 Table 4: Data Information Data Lit1 Lit2 Myst Sci SF 1 2 271 78 272 58 73 11 88 94 56 119 3 67 72 7 62 75 Rating 4 5 6 7 8 9 10 74 106 125 83 70 40 22 92 56 75 104 87 77 42 8 29 46 45 64 66 151 49 51 53 35 31 16 21 83 67 33 28 21 12 6 Table 5: Data Rating Distributions tion." Two separate sets were randomly selected from this dataset, one with 936 books and one with 935, and rated by two di erent users. These sets will be called Lit1 and Lit2, respectively. The remaining sets were based on all of the adequate-information Amazon titles for mystery" 7,285 titles, science" 6,177 titles, and science ction" 3,813 titles. From each of these sets, 500 titles were chosen at random and rated by a user the same user rated both the science and science ction books. These sets will be called Myst, Sci, and SF, respectively. In order to present a quantitative picture of performance on a realistic sample; books to be rated where selected at random. However, this means that many books may not have been familiar to the user, in which case, the user was asked to supply a rating based on reviewing the Amazon page describing the book. Table 4 presents some statistics about the data and Table 5 presents the number of books in each rating category. Note that overall the data sets have quite di erent ratings distributions. 3.1.2 Performance Evaluation To test the system, we performed 10-fold cross-validation, in which each data set is randomly split into 10 equal-size segments and results are averaged over 10 trials, each time leaving a separate segment out for independent testing, and training the system on the remaining data 23 . In order to observe performance given varying amounts of training data, learning curves were generated by testing the system after training on increasing subsets of the overall training data. A number of metrics were used to measure performa...
View Full Document

{[ snackBarMessage ]}