Connecting the dots: Reader ratings, bibliographic data, and machine-learning algorithms for monograph selection
•
1 recomendación•102 vistas
Jingshan Xiao University of Houston Clear Lake Wenli Gao Data Services Librarian, University of Houston Traditional collection development relies heavily on human input, with librarians relying on reviews and subject selection lists, and through user requests. With the development of machine learning, more and more businesses seek automated methods to deliver results relevant to users. The Recommender system, a subclass of information filtering that seeks to predict the "rating" or "preference" of a user, is among the most successful systems of machine learning in action. It has been adopted by many major e-commerce businesses such as Amazon, Netflix, and Expedia, and has been widely implemented to predict product and media recommendations, making it a key factor in increasing product average order value and the number of items per order. Drawing inspiration from the benefits of a recommender system to business and its success in heightening the reliability of recommendations, we attempted to build optimal collection recommendations with machine-learning algorithms using Python. The purpose of this project is to help librarians make collection decisions using the recommender system, and in this presentation we will illustrate several examples of building this system to aid in the selection of monographs. One example involves the merging of popular titles with reader rating data. We found that while The New York Times publishes best seller titles based on the rates of sales, they do not have any connection to user ratings. By leveraging data from Goodreads, the world’s largest site for readers and book recommendations, we will build a simple recommender system that produces The New York Times best seller titles that have higher user rating using a matrix factorization based method. Another example of using a recommender system is to have the ability to refer selectors to books that are similar to a particular title based on pairwise similarity scores. News services are already able to identify related articles of interest to readers based on the articles that they have read in the past, so applying this system to libraries is an exciting prospect. Drawing on bibliographic data from highly circulated items, the recommender system will suggest items with similar features using similarity metrics. The recommender system will use machine-learning algorithms not only to simplify collection development for librarians, but will also help end users discover more items relevant to their interests.
Recomendados
Recomendados
Más contenido relacionado
Más de NASIG
Más de NASIG (20)
Último
Último (20)
Connecting the dots: Reader ratings, bibliographic data, and machine-learning algorithms for monograph selection
Notas del editor
- Goodreads ratings, metadata from oclc, python
- Briefly talk about the evolution of data, concepts of RS
- chronologically or geologically organized. Get all articles from 2007, manageable
- find books on a subject, by Wheeler—easy where he isn’t the first author—not that easy
- With the rapidly increasing generation of data, users experience information overload and face difficulties in navigating and processing the information available online. Human’s capacity to find information advances more slowly than the pace at which new information is made available. More recently, terms such as big data, machine learning, and artificial intelligence no longer appear only in research papers. They appear so often in popular media that the general public have some familiarity with the terms. Big data are high-volume, high velocity, and high variety information assets that require new forms of processing to enable enhanced decision making, insight discovery and process optimization.
- Different library tools generate tons of library big data, mainly catalogue and transactional data. Catalogue data provides metadata information, while transactional data such as circulation statistics offers insights on user behaviors. There exists plenty of opportunities to analyze library big data. The rapid increase in library data offers innovative ways to understand interactions with users in the library environment. Analytics of library big data support library innovations, personalized recommendation services, library user behavior analysis
- ML is the study of algorithms and statistical models that computer systems use to effectively perform a specific task without using explicit instructions, relying on patterns and inference instead. As this recent article says,…Giving enough data, we can train the computer to recognize patterns. It is very powerful. RS is the result of ML of metadata and user behavior data.
- More and more businesses seek automated methods to deliver results relevant to users with the development of machine learning. Recommender systems “analyze user profiles, content items, and the connections between them, and try to predict future user behavior” It has been adopted by many major e-commerce businesses such as Amazon, Netflix, and Expedia, and has been widely implemented to predict product and media recommendations, making it a key factor in increasing product average order value and the number of items per order. Recommender systems lead to improved browser-to-buyer rates, extended cross selling opportunities and increased customer loyalty, while also reducing the time and effort spent by consumers on searching Recommender systems are widely used to suggest contacts, or activities on social media platforms, and to improve targeted ads by the advertising industry Practitioners can also develop new marketing strategies by integrating users’ current situations and future needs by offering contextually relevant socialized recommendations
- 35% of the purchases on Amazon are the result of their recommender system, according to McKinsey. During the Chinese global shopping festival of November 11, 2016, Alibaba achieved growth of up to 20% of their conversion rate using personalized landing pages, according to Alizila. Recommendations are responsible for 70% of the time people spend watching videos on YouTube. 75% of what people are watching on Netflix comes from recommendations, according to McKinsey. Employing a recommender system enables Netflix to save around $1 billion each year—less on ads
- BibTip used for the previous mentioned catalogue
- Hamlet is a tool that uses machine learning to explor MIT thesis collection. Plug in a thesis and find out which other theses are most conceptually similar.
- The ScienceDirect free Recommendations service uses machine learning and your online activity to suggest research tailored to your needs. Once you've signed in, the algorithm uses your signed-in activity on ScienceDirect to understand your research interests. It then searches databases of more than 3,800 journals and over 37,000 book titles to find related content. The more frequently you sign in, the better it gets to know you, and the more relevant the recommendations you'll receive.
- Help library users discover material relevant to their research Predict individual preferences, help users make better choices
- Lu, J., Wu, D., Mao, M., Wang, W., & Zhang, G. (2015). Recommender system application developments: a survey. Decision Support Systems, 74, 12-32.
- Collaborative filtering-based recommendation techniques combine users in some way and creates recommendations for one user based on the preferences of another user. It helps make choices based on the opinions of other people who share similar interests. Content based pair specific users to library items based on the metadata of the item and what is known about the user. For example, if a user indicates in some way that they enjoy mystery novels, items identified as mystery novels might be recommended to them.
- This project is to help librarians make collection decisions using the recommender system, and in this presentation we will illustrate several examples of building this system to aid in the selection of monographs. One example involves the merging of popular titles with reader rating data. We found that while The New York Times publishes best seller titles based on the rates of sales, they do not have any connection to user ratings. By leveraging data from Goodreads, the world’s largest site for readers and book recommendations, we will build a simple recommender system that produces The New York Times best seller titles that have higher user rating. Drawing on bibliographic data from highly rated best sellers, we used a few methods to suggest items with similar features.
- Data for this project was retrieved from APIs. Weekly hardcover fiction best sellers in the year of 2018 was collected using New York Times API. A total of 52 weeks of data, containing 15 best seller titles, was retrieved in CSV files. After combing 52 files and removing duplicated titles, a total of 192 titles comprise the 2018 New York Times best seller books. The file contains fields such as ISBN, title, author, subject and summary, with no review information.
- Goodreads also provides free API to retrieve review data. The project team collected reviews in 2018 by its top 99 reviewers. Among those, 21 had reviewed New York Times best seller titles, with a number of 1367 reviews available. Goodreads review data contains userID, rating, goodreadsID, and title. GoodreadsIDs were matched with ISBN to create a list of New York Times best sellers with rating.
- Bibliometric data came from WorldCat Search API. Unlike the free New York Times API and Goodreads API, WordCat Search API came with WorldCat services subscriber. Data was in XML file format, and Python Pandas library was used to parse the data. author, book summary, title, topic term, and genre
- Pandas for query csv files, numpy for correlation Sklearn for analysis, alternative: spaCy-Natural Language Processing in Python. It features NER, POS tagging, dependency parsing, word vectors and more.
- Bayesian Estimate Algorithm offers generalized recommendations based on item popularity score. The basic idea behind this algorithm is that items that are more popular and critically acclaimed will have a higher probability of being liked by the average users. Using this method, each title will generate a list of books with a similarity score. The higher the score, the more closed related the books are. The values are need to determined an appropriate value for m percentile as cutoff. (more high value more small output)
- The score here is similarity score, weighted rating
- Matrix factorization algorithm is a popular collaborative filtering model, been widely adopted in the recommendation field for a long time. Matrix factorization is the breaking down of one matrix into matrix of users and items. The main idea is to make recommendation in the large collaborative sparse matrix through analyzing two small and low-dimensional matrices, such as reader and books.
- Using SVD. The Singular-Value Decomposition (SVD), is a matrix decomposition method for reducing a matrix in order to make certain subsequent matrix calculations simpler.
- Cosine similarity score is another one of popular content based recommendation techniques. It converts a collection of metadata text to a matrix of token counts, then compute the cosine similarity score for each word. The cosine score can take any value between -1 and 1. The higher the cosine score, the more similar the documents are to each other.
- By calculating cosine similarity score from metadata field author, title, topic term, and genre, a list of recommended book was generated.
- This method is frequently used in natural language processing. It converts a collection of summary to a matrix of TF-IDF that reflect how important a word is to each summary by giving a weight factor. After the calculation of TF-IDF, cosine similarity algorithm is used to measure the similarity for each word.
- There are limitations to recommendation systems. For this project, the project team used leisure reading titles as a pilot. As some recommendation techniques reply on reader reviews, the lack of availability of review activities for academic books makes it hard to use reader reviews to support recommendations for academics. On a larger scale, it raised the concern for privacy. User behavior data are sensitive information. Personalized recommendation service should not sacrifice user’s privacy. Besides, the algorithms that determine which users are similar and thus which recommendations to make are not often understood. Reidsma (2016) pointed out that "in librarianship over the past few decades, the profession has had to grapple with the perception that computers are better at finding relevant information than people." The algorithms that are doing the finding, however, often carry the same hidden biases that their programmers have. Reidsma (2016) encourages a broader understanding of algorithms in general and deeper understanding of recommendation algorithms in particular.
- This projects highlights several ways of recommendations by using New York Times bestseller titles. The technique has potential to be implemented in academic library settings, using content based metadata. Libraries could run reports for highly circulated items and find similar books to those items for collection development. It is the same with interlibrary loan data. Libraries could find out similar items to frequently requested interlibrary loan items to bridge the gap in the collection. Moreover, retrieving data from libraries, especially libraries with similar user profile, could generate cross-library recommendations. The recommender system will use machine-learning algorithms not only to simplify collection development for librarians, but will also help end users discover more items relevant to their interests.