Skip to main content

Hitachi

Corporate InformationResearch & Development

Photo: SUZUKI Masato
SUZUKI Masato
Researcher

Through digital broadcasts, a lot of TV programs can be offered to viewers. But finding the programs one wants to watch with certainty can be troublesome. It is not unusual that a viewer misses a favorite program while being unaware that the program is being broadcast. With that problem in mind, we would like to propose the possibility of asking the TV "Are there any programs I might like on TV right now?" while switching TV channels. To make that possibility a reality, in May 2011, Hitachi launched TVs featuring a function called "Hitachi Suggested Programs" in Japan.

Recommending programs that fit a user's tastes

What kind of function is "Hitachi Suggested Programs"?

Photo: SUZUKI Masato

SUZUKIIn a few words, it is a function—which works on the TV set itself—for recommending TV programs. In accordance with recording and playbacking histories of user, the tastes of the viewer are analyzed, and programs that fit those tastes are recommended to the user.

As for the number of digital-broadcast programs, if all terrestrial digital broadcasts and BS (broadcast satellite) digital programs in the Tokyo metropolitan areas are taken together, around 8000 programs are broadcast per week. Searching for programs that a viewer wants to watch from among those 8000 is troublesome. As conventional procedures for searching for programs, "view program guide," "use search function," and so on are available; however, they always involve searching for target programs from the viewer's side. And sometimes a viewer feels "I want to watch something good even though I don't have a particular target program." Considering that a function that suggests programs from the TV side at such times would be useful, we have developed such a function. Consecutively switching from channel to channel while thinking "there must be a good program on somewhere" is called "zapping," this function provides a means of searching for programs with the same kind of intuition.

How does this function learn the tastes of a viewer?

SUZUKIIn the case of digital broadcasts, program guides are transmitted from broadcasting stations. Information such as scheduled times, program titles, and genres included in those program guides is used to analyze the tastes of a viewer. For example, for people who often watch TV dramas at 9pm on Mondays, dramas broadcast at 9pm on Mondays are recommended on the basis of information stated as "drama", "9pm", and "Monday".

The information received from the broadcast station, however, is not sufficient for the analysis of viewer tastes. For example, dramas are categorized only as domestic and foreign. And dramas categorized as foreign include those from a variety of countries. Accordingly, as for a way of mostly viewing dramas from specific countries, we have contrived a way to re-categorize data obtained from broadcasters so that dramas from those countries can be mostly recommended.

What's more, program information of digital broadcasts contains program contents and a summary of performers. Information such as performers and director is extracted from this information and used in the analysis of viewer tastes.

Figure 1: A screen shot of the "Hitachi Suggested Programs" function

*
"Rovi", "G-GUIDE", and any other "G-GUIDE" accompanying logos are registered trademarks of Rovi Corporation or its subsidiaries in the USA and in other countries.

Recommendation based on three criteria

What are the criteria for determining what programs to recommend?

SUZUKI Tastes of the viewer are evaluated according to three criteria: certainty, discoverability, and unpredictability. The first criterion, certainty, is the domain in which the viewer would definitely act. For example, it contains programs viewed on a routine basis. If programs in this domain are recommended, it is fairly certain that the viewer will record them and watch them. The second criterion is discoverability. Since a viewer's enjoyment is not assured by only recommending programs that the viewer will watch with certainty, although programs such as variety shows featuring performers liked by the viewer are within the predicted range, programs that the viewer might not know are also recommended. The third criterion is unpredictability. It is the domain in which the viewer will think with an element of surprise: "That's interesting." Unlike programs watched in the usual manner, programs of unimagined interest are recommended.

With the mindset that nobody will pay attention to recommendations that don't fit their tastes, but if the recommendations are too obvious, the viewer will get tired, we considered what would be the recommendation range that would keep the viewer happy. With that question in mind, we decided to determine what programs to recommend on the basis of the three above-described criteria. Considering the process for selecting the programs, we looked up and investigated a certain use case dealing with the question, "For a viewer who often watches dramas featuring celebrity A, what is certain, what is unpredictable, and what is discoverable?"

Figure 2: Analysis of tastes of users (viewers)

What amount of behavioral history is required by the analysis of tastes?

SUZUKIEven if information on only one preferred program is available, programs can be recommended by the Hitachi Suggested Programs function. When the function is started up, the user first registers the programs that he or she likes. The initial analysis is then performed under the conditions set by the registered information, and the recommendation results are presented.

We thought that it is necessary to provide convenience in terms of immediate use when the user decides to use the Hitachi Suggested Programs function for the first time. Given that requirement, we made it possible to present recommended programs on the basis of a few conditions. As another idea to enable immediate use, an algorithm for analyzing tastes at higher speeds than possible with usual recommendation algorithms is utilized to ensure that the recommendation results are presented without making the viewer wait.

Pursuing recommendation results that can satisfy the customer

Did you have any trouble when improving the accuracy of the analysis?

Photo: SUZUKI Masato

SUZUKIYes, we had trouble regarding two points: What information to use for the analysis and what weighting to give that information.

First, I'll talk about the information used. The information extracted from recording and playbacking histories affects the search results. So selecting that information was a process of trial and error. Moreover, since the information obtained from broadcaster is used, we have struggled to extract and narrow down that information.

Next, I'll explain the second point, namely, weighting. As I mentioned above, we cited and considered a use case. And for eliciting the optimum recommendations in regard to that use case, a "weight" was added to each piece of information. In accordance with the balance of the weighting, the programs being recommended change. For example, as tastes of a certain person, two keywords—"drama" and "entertainer A"—are taken at similar number of times. If weighting of each keyword is same, in addition to dramas featuring entertainer A being recommended, dramas unrelated to entertainer A as well as variety shows and quizzes featuring entertainer A are recommended. However, if "performer" is weighted as 1 and "genre" is weighted as 10, most recommended programs would be dramas. As a result, as for programs related to entertainer A, programs other than dramas will hardly be recommended. What weight to give should be that which satisfies the user. Investigating that weight and adjusting it was a tough task.

You investigated a lot of use cases, didn't you?

SUZUKIYes, we did. To put a certain use case into practice, if extracted information and its weighting are investigated, consistency with weightings and information extracted with other use cases is required, so the investigation continued for over a month. We made great efforts, and despite coming up with a tentative theory and testing it, we often got results that were completely beyond our expectations. And on getting such astounding results, we laughed out loud.

You were also involved with the way of displaying recommendations on an actual screen, right?

SUZUKIWe asked designers to incorporate our ideas about how to present recommended programs to viewers in an easy-to-follow manner. For example, the "recommendation level" is presented to viewer in the form of a number of stars. Members of our research team expressed the opinion that in the case of simply listing the recommendations, the viewer wouldn't understand to what extent programs were being recommended. With that opinion in mind, we decided to list the recommendation in accordance with the recommendation level.

"Viewing the desired contents at the desired time in the desired manner"

Tell us about the features of Hitachi TVs in addition to the Hitachi Suggested Programs feature.

SUZUKIHitachi believes that the TV should enable viewers to "watch the desired contents at the desired time in the desired manner." We believe that capability is perfectly represented by Hitachi's TVs launched up till now.

For example, providing the "home network function compliant to DLNA" and having compatibility with the iVDR-S storage medium (namely, a cassette-type HDD) make it possible to view recorded programs at the desired time. Programs can be recorded onto iVDR-S directly, and programs recorded on a TV's built-in HDD can be dubbed onto iVDR-S. Data recorded on iVDR-S can be carried to other rooms in the home and played on other TVs and iVDR-S-compatible devices in the same manner as video cassettes and DVDs. Moreover, the home network function compliant to DLNA can be used to replay programs via a network. In other words, the way to view the program contents can be chosen according to the viewer's desire.

In addition, a function called "auto-chapter" for automatically creating chapters at the highlight points of programs is provided. For example, during a music show, it is possible to present songs that the viewer wants to see by jumping to the start of the songs. In line with the above-mentioned theme, namely, "watching the desired contents at the desired time in the desired manner," various features are implemented in Hitachi's TV.

*
iVDR is a trademark indicating compliance with the iVDR technology standard.
*
iVDR-S indicates inclusion of the SAFIA (security architecture for intelligent attachment device) content protection method.
*
DLNA is a registered trademark of the Digital Living Network Alliance.

Seeking viewer happiness

As for the future of TV, what kind of dreams do you have?

Photo: SUZUKI Masato

SUZUKIActually, from the day I joined Hitachi, I have wanted to realize the same slogan, namely, "watching the desired contents at the desired time in the desired manner." An example is carrying recorded TV programs about and watching them on a mobile phone while commuting. And if only part of a program could be watched, the rest of it can be watched on the TV after returning home. At that time, the environment itself in which to watch videos on a mobile phone had not been set up; however, it has now been set up, but few people use it. These days, even though it is said that the number of people watching TV is decreasing, I think that if technology making it possible to seamlessly carry video were created, an "enjoyable TV lifestyle" would be realized.

What's more, operability that is easily understood from the viewer's viewpoint is essential. Functions that a normal person finds difficult to use are still in great evidence. So my goal is "easy, viewing anytime and anywhere."

What other dreams do you have?

SUZUKISince the TV is something that is used by everyone in a household, I think it would be beneficial to realize "personalization"—namely, the ability to present information from the TV side in accordance with the tastes of each member of the household. Examples of personalization would be presenting cartoons to children, and presenting news programs to adults. In regard to program-recommendation technology as well, I want to realize personalization. Actually, although the opinion that recommended programs should be presented separately in accordance with each user's tastes has been expressed by our researchers, we have not yet made it possible to "automatically present just the right content for each viewer." Realizing that possibility is a challenge we will tackle from now onwards.

You are still chasing user happiness, right.

SUZUKIYes, I am. I am in charge of developing so-called consumer-oriented products. On the one hand, I am a producer; on the other hand, I am a user. That said, I want to pursue development from both viewpoints. Since I joined Hitachi, my superiors have advised me to "make things that you would want to buy yourself." So I am pushing ahead with my research in accordance with that advice.

(Publication: July 9, 2012)

Notification

  • Publication: July 9, 2012
  • Professional affiliation and official position are at the time of publication.

  • Page top

Related contents

  • Page top