For broadcasters that have entered the era of file-based production, there is a new range of possibilities for content distribution to suit new demands from their viewers. In the early days, programs were created for the classical linear TV broadcast. For example, a news episode was typically about 30 minutes long, with about 10 internal affairs, five external affairs and five sports items. Viewers watched the news items in the order the editor-in-chief had decided.
The Internet changed this lean-back approach. Broadcasters make individual items available on the Web, allowing consumers to select and view only those items that interest them. File-based production is an enabling technology as it simplified and facilitated this transition.
But this is only the tip of the iceberg. This article will present another promising application made possible with file-based production and its associated well-structured metadata — an application that won't create an additional burden on the existing production processes.
The Internet makes news consumption flexible, but it also requires consumers to be more active. In particular, they have to select items manually — a lean-forward approach. Some technologies, such as Really Simple Syndication (RSS), promised an automated delivery of items, yet these are typically limited to deliver items in a certain category, such as sports items.
Although the Internet made it possible to distribute more items than in a classical TV newscast, broadcasters discard a lot of material — for example, clips received from international content providers such as the EBU/Eurovision or Reuters. Material is discarded not only due to a lack of editors, but also it might be deemed as of interest for a too small group of consumers. American football, for instance, is not a popular sport in Belgium; hence, almost no items about American football are available through mainstream broadcasters.
This article will illustrate that by analyzing and combining the existing metadata in a file-based news production workflow, broadcasters can automatically generate custom-made news episodes from all available content and at the same time re-enable a lean-back consumption style. This article will use as a model the system at Vlaamse Radio-en Televisieomroep (VRT), the public service broadcaster of the Flemish part of Belgium.
Personalization
To address viewers' individual preferences for news in the context of TV broadcasts (fixed sequence of items and duration) and Internet distribution (lean-forward approach), news consumers should have the ability to personalize the news by being able to:
- Select only the news categories they are interested in — for example, local political news and international business news, but no international sports items.
- Sort the news categories according to their personal preferences — for example, first business, then politics, and then science and technology.
- Choose to receive all news material on a particular topic, as well as those topics deemed not interesting for a large audience, such as all available news items about science and technology.
- Specify the maximum duration of the news. For example, a viewer's commute time is 30 minutes, so he wants to receive no more than 30 minutes of news material.
- Watch the news items offline so they can consume the news on a mobile device.
Once consumers' preferences are known, up-to-date news bulletins can be created and distributed fully automated, resulting in a personalized and lean-back consumption style.
Goals and requirements
VRT's goal was to create a solution that automates the composition and distribution of personalized news without imposing any additional burden on the existing production processes. The composition creation process takes the user preferences and conditions listed in the previous section and the available news items into account to digest an up-to-date news bulletin.
The news bulletin is distributed by using vodcast technology. A vodcast is a variant of a podcast, which on its turn is an RSS feed whereby the items can refer to a nonembedded video fragment. Optimized vodcast players, such as Apple's iTunes, retrieve the necessary video files and store them locally for playout and offline consumption. Each time users synchronize the vodcast, e.g., on their mobile devices, they automatically receive personal up-to-date news on that device. Figure 1 displays the overall architecture of the solution. It is divided into a back-end and a front-end.
Back-end: Newsgathering
The back-end takes care of the retrieval, conversion and storage of any news material (i.e., metadata and audiovisual content) so it is available for the front-end. It collects information from two sources.
First, it collects information from VRT's Avid iNews digital newsroom system after the broadcast of a regular TV news bulletin. The related audiovisual material is retrieved from the production Vizrt Ardome by a media asset management (MAM) system.
Secondly, news clips are provided by the content provider EBU/Eurovision. The audiovisual material is retrieved from EBU's SuperPOP (point of presence) news distribution system developed by Thomson Grass Valley and transcoded to Material eXchange Format (MXF) files containing DV25 encoded video and uncompressed PCM audio. The metadata is fetched as NewsML-G2 descriptions from an EBU Web server. (IPTC's NewsML-G2 standardizes how to structure and specify metadata associated to news stories.)
Note that only EBU/Eurovision is currently integrated as this content provider already describes its content using the NewsML-G2 specification, which makes automated processing possible. Adding other content providers to the architecture is straightforward if NewsML-G2 is used or if the format used can be converted to NewsML-G2. For example, iNews generates a collection of Standard Generalized Markup Language (SGML) files from which the relevant information is extracted and combined into NewsML-G2 compliant files.
All relevant information contained in the NewsML-G2 description — such as title, content description, rights information, publication date, owner name, media file name and location — and the audiovisual content files are fed into the central MAM system using Simple Object Access Protocol (SOAP). Figure 2 displays an item in the MAM system received from EBU/Eurovision.
Front-end: Personalization and distribution
The front-end takes care of the news composition and distribution. First, the information in the back-end is checked for new or updated items. These items are combined in such a way to suppress duplicates and identify relations between the items. The video material is retrieved from the MAM system and converted to a consumer playable format, namely MPEG-4 AVC/H.264. This format was selected because it can be used on handheld devices such as Apple's iPod and Windows Mobile 6.1-based machines.
Second, a Web interface, illustrated in Figure 3, allows news consumers to specify their preferences and preconditions. Consumers can order the news categories to their liking and specify if they want to retrieve the nonbroadcasted material from EBU/Eurovision belonging to a particular category as well. Additionally, users can specify their available time span, meaning that the total duration of the material in the resulting personalized news will not exceed this. The result of this configuration Web page is a particular crafted URL containing the necessary information for the vodcast generation. As long as their preferences do not change, they can keep using the same crafted URL without returning to the Web interface.
Third, the front-end generates the vodcast to distribute the up-to-date personalized news. It takes the cached news items and the particular crafted URL as input. In five steps, it creates the personalized news bulletin as a vodcast:
- From all available items, only the items belonging to a category selected by the news consumer are kept.
- By analyzing the versioning information, the algorithm retains the most recent version of the items.
- The remaining items are sorted according to the news consumer's preference.
- If the consumer specified a maximum duration, the algorithm adds items one by one as long as the total duration does not go beyond the specified duration. For example, if the maximum duration is 15 minutes and the total duration of retained items so far is 12 minutes, then the next item of three minutes or less is added to the result list.
- Remaining items are serialized as vodcast items and returned to the news consumer. This vodcast can be processed by any compliant application or device. Figure 4 (below) displays a vodcast in the Firefox Web browser. Clicking on a title starts playing the associated MPEG-4 AVC/H.264 media file.
Each time the feed is synchronized, a new up-to-date vodcast is generated by the same five steps. The application receiving the feed automatically detects the new available items and presents them to the end-users. For instance, Apple's iTunes will retrieve the vodcast at the user's predefined interval (or on request). It detects which items are added, and it retrieves the associated media files. Users can view the textual description and the audiovisual material of these items either in iTunes and (offline) on their iPods.
Conclusion
As file-based production is becoming more and more widespread in the broadcasting community, new applications are appearing that exploit its merits. This article has presented an architecture and an application that illustrates one of the many new applications that file-based production workflows can support without additional effort for broadcasters but with added value for end users.
By intelligently handling the metadata that is available during the news production process, VRT has created an application that allows end users — news consumers — to create personalized news bulletins. Consumers can select and sort their preferred news categories, choose to receive nondistributed material provided by international news agencies, and limit the total duration of their news. As such, broadcasters can offer more content to end users, who can consume this in a lean-backwards way.
VRT's application automatically processes the information from the rundowns of broadcasted TV news bulletins and the NewsML-G2 formatted descriptions of content provider EBU/Eurovision. This article discussed the process to compose the personalized news and how to distribute it as a vodcast so that end users can consume the news online and offline on a mobile device.
One pending issue needs to be resolved before the application can be rolled out in a production environment, namely the processing of the usage and distribution rights of the items. Currently, the NewsML-G2 description format has only basic (textual) provisions to describe the usage rights. As the processing of these textual descriptions cannot be automated, the application currently assumes all usage rights are cleared for distribution. Some additional research is required on how to describe the rights information and how to automate its processing.
Robbie De Sutter is a scientific researcher at VRT-medialab, the R&D department of VRT.
