In this paper, we evaluate the accuracy of personality-based recommendations using a real-world data set from Amazon.com. We automatically infer the personality traits, needs, and values of users based on unstructured user-generated content in social media, rather than administering questionnaires or explicitly asking the users to self-report their characteristics. We nd that personality characteristics signi cantly increase the performance of recommender systems, in general, while di erent personality models exhibit statistically signi cant di erences in predictive performance.
Personality traits have been found to in uence various aspects of individual behavior, including job performance [5], academic motivation [17], and romantic relationships [25]. Despite the initial promising evidence in various academic elds and applications, including recommender systems (RSes), personality traits are still not frequently used in predictive modeling, mainly because they usually require users to complete long questionnaires and hence they cannot be easily applied at a large scale. In this study, we automatically infer cognitive and social characteristics of users based on di erent personality models in psychology, including Big Five, Values, and Needs, and present a comparative analysis.
Tapping into the recent advances of data mining, various
studies have successfully attempted to automatically derive
personality traits from text based on the established
relationship between word use and personality [
In RSes, the use of personality traits is a promising but under-explored research direction. Among the most relevant works, [10, 15] explicitly measure users' personality based on quizzes aiming at alleviating the cold-start problem. Using also questionnaires, [13] nds correlations between personality and movie preferences, while [6] studies the relationship between personality and preferences in multiple entertainment domains using explicit psychometric tests. There are several characteristics though that di erentiate this study from the related work. For instance, apart from the Big Five model [9, 20] that the aforementioned studies employ, we also use the personality models of needs [12, 18] and values [22]. Besides, rather than administering questionnaires or explicitly asking the users to self-report their characteristics as in previous studies in RSes, we automatically infer the personality characteristics, needs, and values of users based on unstructured user-generated content in social media. 3.
The personality traits [9, 20], needs [12, 18], and values
[22] of the users in this study are automatically inferred
based on a textual analysis of user-generated unstructured
data. In particular, for each user we analyzed the content of
all the messages that there were publicly posted over time
on the social network of Twitter as well as the user-de ned
description of their accounts. From the messages of the users
analyzed are excluded all the private messages between the
users as well as non-English messages. In addition, we
excluded any messages that were not written by the speci c
target user each time (e.g., re-tweets) as those messages do
not correspond to the linguistic style of the speci c user
and hence might not re ect her/his personality. After the
pre-processing of the corpus of user-generated content, there
were on average 26; 568 words per user; this number is much
higher than the typical number of words in other studies
(e.g., [13]) and can lead to more accurate results. The
messages and the rest of the user-generated of each target user
are merged into a single \document" and the personality
traits, intrinsic needs, and values of individuals are then
derived using linguistic analytics. In particular, the tokens
of the user-generated content -after some pre-processing of
the words, which includes removal of stop-words and
nonEnglish words, stemming, and fuzzy matching- are matched
with the Linguistic Inquiry and Word Count (LIWC)
psycholinguistic dictionary, which has been developed over
several years and currently includes almost 4,500 words and
word-stems associated with one or more personality
categories [21], to compute relative scores in each dictionary
category. Afterwards, based on [
To empirically evaluate the employed approach, we build a factorization model incorporating the information of personality traits, needs, and values as well as item attributes.
In particular, the user preferences are modeled as: y(x) = y(u; i; 1u; : : : ;
um; 1i; : : : ; ni) = w0 + wu + wi + wj ju +
wl li + hvu; vii n X l=1 m + X j=1 m X j=1 n X l=1 juhvju; vii + lihvu; vl i + i ju lihvju; vlii; where the input vector x 2 RjUj+jIj+m+n contains binary indicators for the user and item, the user attributes u = ( 1u; : : : ; mu) capturing the personality characteristics of the users, and item attributes i = ( 1i; : : : ; mi) capturing the item categories, prices, etc.; the factorization of users vu, items vi, and attributes vju; vli is of dimensionality k.
Our data set was collected as in [1, 24] and contains 906; 277 purchases of 138; 536 distinct products on Amazon.com from 81; 475 users who shared their purchases on Twitter as well as the account information and the usergenerated content on the social network of Twitter for the same users. As our data set includes only implicit ratings, for each user we randomly select an equal number of nonrated items (based on the frequency of ratings of each item) as negative examples in order to increase the accuracy of our predictions. We use MCMC inference with Gibbs sampling to learn our factorization model. Moreover, we employ a holdout evaluation scheme with 80=20 random splits into training and test sets without ltering any ratings and we evaluate each model in term of classi cation performance based on accuracy.
Figure 1 shows the experimental results. We see that personality characteristics increase the performance of RSes and that di erent personality models can result in di erent predictive accuracy. Interestingly, the under-explored personality models of needs [12, 18] and values [22] resulted in better predictive performance compared to the more popular model of Big Five traits [9, 20]. We also see that combining the attributes of the di erent personality models results in even better performance and, hence, has the potential to further increase the business value of recommendations [2, 3].
In this study, we automatically infer the personality traits, needs, and values of users based on unstructured user-generated content in social media and build di erent RS models. Using m n X X data from Amazon.com, we nd that personality characteristics can increase the performance of RSes and we identify a speci c model of personality that signi cantly outperforms the remaining models achieving promising performance.
The main advantage of the employed approach is that automated methods for personality assessment are more efcient and objective [11]. In particular, the traditional way of measuring personality, which requires people to complete long questionnaires, does not allow to obtain personality traits at a large scale for the population of interest [7]. Besides, user-generated content is more re ective of users' actual personalities, not \idealized" versions of themselves [4]. 6. [5] Barrick, M., and Mount, M. The big ve personality dimensions and job performance: A meta-analysis. Pers.
Psychol. (1991). [7] Chen, J., et al. Making use of derived personality: The case of social media ad targeting. In ICWSM (2015), AAAI. [17] Komarraju, M., and Karau, S. J. The relationship between the big ve personality traits and academic motivation. Pers. individual di erences39, 3 (2005), 557{567. [19] Mairesse, F., and Walker, M. Words mark the nerds: Computational models of personality recognition through language. In CogSci (2006). [20] Norman, W. Toward an adequate taxonomy of personality attributes: Replicated factor structure in peer nomination personality ratings. The J. Abnorm. Soc. Psychol.66, 6 (1963). [22] Schwartz, S. H. Basic human values: Theory, measurement, and applications. Revue francaise de sociologie47, 4 (2006). [23] Tausczik, Y., and Pennebaker, J. The psychological meaning of words: Liwc and computerized text analysis methods. J.
language social psychology29, 1 (2010), 24{54. [24] Todri, V., and Adamopoulos, P. Social commerce: An empirical examination of the antecedents and consequences of commerce in social network platforms. In ICIS (2014). [25] Tupes, E., and Christal, R. Recurrent personality factors based on trait ratings. J. personality60, 2 (1992), 225{251. [26] Yang, H., and Li, Y. Identifying user needs from social media.
Tech. rep., IBM Tech Report. goo. gl/2XB7NY, 2013.