This paper presents a content-based book recommendation system called LIBRA, which uses information extraction and a machine learning algorithm for text categorization. Unlike collaborative filtering, which relies on user preferences, content-based methods use information about the item itself to make recommendations. LIBRA extracts book information from Amazon.com and uses a Bayesian learning algorithm to learn user profiles based on user ratings of training books. The system then recommends additional titles based on the user's preferences. The system first extracts information from Amazon.com pages using a pattern-based information extraction system. This information is then processed into a vector of bags of words for each slot (e.g., title, authors, synopses, etc.). The user then rates a set of training books, and the system learns a profile of the user based on these ratings. The system then recommends additional titles based on the user's preferences. The system's performance is evaluated using several metrics, including classification accuracy, recall, precision, and rank correlation. The results show that the system performs well, even with small training sets. The system also has the ability to explain its recommendations by listing the features that contributed to its high rank. The paper also discusses the role of collaborative content in content-based recommending. An ablation study was conducted to determine whether collaborative content actually helps the system's performance. The results indicate that collaborative content has a significant positive effect on the system's performance. Future work includes developing a web-based interface for LIBRA, comparing its content-based approach to a standard collaborative method, and exploring additional ways of combining content-based and collaborative recommending. The paper concludes that content-based recommending has the potential to provide accurate and personalized recommendations, even in the absence of information about other users.This paper presents a content-based book recommendation system called LIBRA, which uses information extraction and a machine learning algorithm for text categorization. Unlike collaborative filtering, which relies on user preferences, content-based methods use information about the item itself to make recommendations. LIBRA extracts book information from Amazon.com and uses a Bayesian learning algorithm to learn user profiles based on user ratings of training books. The system then recommends additional titles based on the user's preferences. The system first extracts information from Amazon.com pages using a pattern-based information extraction system. This information is then processed into a vector of bags of words for each slot (e.g., title, authors, synopses, etc.). The user then rates a set of training books, and the system learns a profile of the user based on these ratings. The system then recommends additional titles based on the user's preferences. The system's performance is evaluated using several metrics, including classification accuracy, recall, precision, and rank correlation. The results show that the system performs well, even with small training sets. The system also has the ability to explain its recommendations by listing the features that contributed to its high rank. The paper also discusses the role of collaborative content in content-based recommending. An ablation study was conducted to determine whether collaborative content actually helps the system's performance. The results indicate that collaborative content has a significant positive effect on the system's performance. Future work includes developing a web-based interface for LIBRA, comparing its content-based approach to a standard collaborative method, and exploring additional ways of combining content-based and collaborative recommending. The paper concludes that content-based recommending has the potential to provide accurate and personalized recommendations, even in the absence of information about other users.