Guidelines for adaptive content generation for television that enhance the viewer experience

Abstract

Keywords

Sentiment analysis adaptive content interactive television social network machine learning real time

On July 8, 2017 at 5 pm, Victor Manuel Mondragón Maca defended his Ph.D. thesis entitled “Guidelines for adaptive content generation for television that enhance the viewer experience” at the University of Oviedo. Victor Manuel Mondragón Maca presented his dissertation in a publicly open event held in the School of Computer Engineering, and was able to defend all his work on every question raised by his thesis committee and the audience. The thesis was supervised by his advisors, Vicente García-Díaz and Edward Rolando Núñez-Valdéz, together with the thesis committee, Juan Luis Pavón Mestras, Juan Manuel Cueva Lovelle and Oscar Sanjuan Martínez. It has been approved, receiving the highest rating. All the cited people were present at the event.

Fig. 1.

Proposal overview.

1. Thesis summary

The technological evolution has promoted changes on how people watch television, causing a metamorphosis of traditional television towards an interactive television (ITV). That, along with the expansion of social networks have promoted the active participation of the viewer. In consequence, there is a need to create audiovisual interactive content that improves the experience of viewers, so that, this Ph.D. thesis aims to present guidelines for the development of audiovisual content with smart adaptability in ITV by using social networks as feedback and interaction channel with the viewer. The proposal structures the research development in three components:

First, the theoretical framework, where a set of ideas, procedures and theories are proposed to support the research done, in addition to including a new definition of the interactive dimension and social resonance applied in the ITV environment.

Second, audiovisual content with smart adaptability in ITV, that constitutes an innovative proposal, where the audiovisual content is able to change its script and scenes in real time to fulfill the expectations of viewers, according to the criteria of the production or direction team.

And third, a proposal of a predictive model in ITV to analyze and predict the sentiment of the viewer about the audiovisual content. The development of this model requires to obtain the opinions the viewers post on social networks, as well as to apply techniques for the analysis of sentiment and to consolidate the machine learning algorithm for the predictive analysis.

To carry out the study, real TV shows were used in Colombia as test pilots to prove the hypothesis of the Ph.D. thesis. Figure 1 shows a summary of the procedure. The main steps are the following [5]:

The content is broadcasted to the network and visualized by the end users.

The viewers watch the show and publish their opinions on social networks.

The opinions and expressions of the users are obtained using different interfaces, specific to each social network.

The data is collected and stored in a repository to be processed using different machine learning-based models, tuned with different parameters based on our previous work [3]. The model is trained to process the subjective sentiments that are expressed by users as published in Mondragón et al. [4].

The model is used to predict the preferences of the viewers over present and future broadcasted content applying the concept of adaptive and smart content.

The main contribution of the work is to correlate the result of the analysis of sentiments done by the model with variables of influence in the historic or present daily thoughts of the viewers to obtain information about trends of collective thinking that can be applied for future broadcasting. That way, the system works as a content recommender system for users on the basis of some implicit feedback techniques [2] that we also applied in recommendation systems for smart electronic books [1], although we focused on smart interactive television content for this research work instead.

References

E.R.

Núñez-Valdéz,

J.M.

Cueva Lovelle,

Infante Hernández,

A.J.

Fuente and

J.E.

Labra-Gayo, Creating recommendations on electronic books: A collaborative learning implicit approach, Computers in Human Behavior 51 (2015), 1320–1330. doi:10.1016/j.chb.2014.10.057.

E.R.

Núñez-Valdéz,

J.M.

Cueva Lovelle,

Sanjuán Martínez,

García-Díaz,

Ordóñez de Pablos and

C.E.

Montenegro Marín, Implicit feedback techniques on recommender systems applied to electronic books, Computers in Human Behavior 28(4) (2012), 1186–1193. doi:10.1016/j.chb.2012.02.001.

García-Díaz,

J.P.

Espada,

González Crespo,

B.C.

Pelayo G-Bustelo and

J.M.

Cueva Lovelle, An approach to improve the accuracy of probabilistic classifiers for decision support systems in sentiment analysis, Applied Soft Computing (First Online) (2017).

V.M.

Mondragon,

García-Díaz,

Jordán and

V.B.

Semwal, Measurement of viewer sentiment to improve the quality of television and interactive content using adaptive content, in: International Conference on Electrical, Electronics, and Optimization Techniques (ICEEOT), IEEE, 2016, pp. 4445–4450. doi:10.1109/ICEEOT.2016.7755559.

V.M.

Mondragon,

García-Díaz,

Porcel and

González Crespo, Adaptive contents for interactive TV guided by machine learning based on predictive sentiment analysis of data, Soft Computing, First Online (2017), 1–22.