In this paper we present a time-based genre prediction strategy that can inform the book recommendation process. To explicitly consider time in predicting genres of interest, we rely on a popular time series forecasting model as well as reading patterns of each individual reader or group of readers (in case of libraries or publishing companies). Based on a conducted initial assessment using the Amazon dataset, we demonstrate our strategy outperforms its baseline counterpart.
Books, which constitute a billion dollar industry1, are the
most popular reading material among all generations of
readers, both for leisure and educational purposes. Hundreds of
thousands of books of di erent types (e.g., paperback and
ebooks) and styles (e.g., ction and non- ction) are published
on a yearly basis, giving readers a variety of options to choose
from. Book recommendation systems, which are meant to
enhance the decision making process, can help users by
identifying, among the sometimes overwhelming number of diverse
books, the ones that best suit their interests and preferences.
These recommenders are not exclusively designed to aid
individuals in their quest for reading materials. They can also
improve the decision making process for libraries, by
suggesting what books to buy in order to maximize the use of
library resources by their patrons, and publishing companies,
by advising which books to publish in order to maximize
revenue. To better serve stakeholders, recommenders must
be able to predict interest and needs. However, given that
preferences may alter over time for di erent readers, the
1http://goo.gl/GMn8Nc
Copyright held by the author(s).
time component is important and crucial to consider in the
prediction process [
There are many avenues that can be explored from a
timesensitive stand point in order to generate better
recommendations, including user-generated ratings, reviews or books
metadata. One of them, which is often overlooked and is the
focus of our paper, is genre. By its de nition, genre (e.g.,
drama, comedy) is a category of literary composition,
determined by literary technique, tone, content, or even length.
While genre has been studied as a part of the
recommendation process [
From the users' point of view, explicitly including time based analysis to inform the recommendation process will lead to relevant suggestions that satisfy their speci c reading needs. Finally, from the commercial point of view, the bene t would be in understanding the in uence of reading patterns on decisions about what genre should be published or acquired in a given point of time. 2.
A considerable number of studies examine the importance
of book genre on readers' activity [
Predicting genre to inform the recommendation process, regardless of the major stakeholder (a reader, library or publishing company), involves examining genres considered in the past, either read by speci c users or purchased by customers. While a simple genre distribution analysis yields probabilities or weights that determine the most favored genres, it lacks the ability to consider genre preference evolution over time. To overcome this drawback, we propose a time-based genre examination which requires information on reading activities among readers. As a rst step to our proposed strategy, we explore reading activity of a user to obtain the distribution of his/her genre interest during continuous periods of time. IN every period of time, for each genre we calculate a signi cance score that captures its importance by considering a number of books read of that genre in that period of time. Thereafter, to explicitly consider the change of genre preference distribution over time, our genre prediction strategy takes advantage of Auto-Regressive Integrated Moving Average2 (ARIMA). We selected ARIMA since it is one of the most popular models that uses time series for prediction purposes.
By using ARIMA we are able to determine a model tailored to each genre distribution to predict its importance for the corresponding user in real time based on its previous occurrences. Note that each predicted genre importance score is based on: its occurrences in the past, a speci c time when it occurred and its importance for a speci c user. To de ne length of time periods used by ARIMA, we used information from a recent study done by Pew3 on reading habits in the USA, we establish one month long \windows" of time in which each user is expected to read at least one book, so our strategy uses 1 month time frames from the the rst book log (either bookmarked or rated book) to last.
Framework. To validate the performance of our proposed
time-based genre prediction strategy, we selected a subset of
the Amazon/LibraryThing4 book dataset. Since the dataset
does not always include genre as a part of the provided
metadata, we extended it by including genre information from
the Library of Congress5. We used 1214 users6 along with
the books they rated or reviewed. To quantify the
assessment, we applied Mean Average Error (MAE), Accuracy and
Kullback-Leibler (KL) divergence [
Results. As shown in Table 1, for N=117 outperform the baseline. KL divergence scores showcase that genre distribution predicted using time-series approach better approx2http://goo.gl/Dhzcg7 3http://goo.gl/BAUQK4 4http://goo.gl/drH0yF 5https://www.loc.gov/ 6In our initial assessment, we considered Amazon users who provided ratings for at least 35 books. 7We empirically veri ed that for 6<N<11 the results are comparable to the ones for N=11 With Time Series Without Time Series With Time Series (3+ genre) Without Time Series (3+ genre) imates to the ground truth. Furthermore, the probability of occurrence of each considered genre is closer to the real values when the time component is included in the prediction process. As a further assessment, we observed the di erences in genre predictions among users who read di erent number of distinct genres. For users who read only one to two genres, the time-based prediction strategy does not perform better than the baseline. However, if a user reads three or more genres, our time-based genre prediction strategy outperforms the baseline in all three metrics. This is not surprising, given that it is not hard to determine area(s) of interest for a user who constantly reads only one or two book genres, which is why the baseline performs as good as time-based prediction strategy. Given that users that read 3 or more genres represent 91% of the users in our sampled dataset, the proposed strategy provides signi cant improvements in predicting preferred genre for the vast majority of readers.
In this paper, we described our e orts in developing a time-based genre prediction strategy that can better inform the recommendation process. The novelty of our approach consists of incorporating an explicit time component to generate genre distribution. To the best of our knowledge, this is the rst time that the well-known time series ARIMA model is used to predict book genre of readers' interests. The described strategy provides successful predictions and outperforms the baseline for 77% of users based on the presented initial evaluation, while for the remaining users it provides predictions comparable to the baseline. Because of the scope of this paper, the conducted evaluation showcases the genre prediction performance for a single user, while we still need to conduct further assessments in terms of quantitatively determining the degree to which the proposed strategy (i)provides successful genre predictions for libraries and (ii)publishing companies and in uences the recommendation process to assist all three stakeholders.