Developing the next generation of multidimensional recommendations


Challenge identifier: DPC1-2018


Proposed by

Altice Labs S.A., an Altice Portugal Group company for innovation, developing new telecommunication solutions and new technologies.



Television has evolved greatly in recent years. Each day more channels, more content, and more interactive services are available to clients. Without an easy and enjoyable way to make a choice of what to see and when to see it, this multitude of content becomes a disadvantage rather than an advantage. Easing the selection process and focusing on the personal interest of each user is a benefit to any TV provider.

This has been an on-going data science challenge to effectively target viewers for a decade. However, it  now has added complexity due to the sheer volume of competing options and the variety of device through which television content can be consumed. So as IPTV services may also be provided through different devices, the content choice is also impacted by the specific consumption context – for example, the same person may have different consumption behaviour depending at the micro level on the device and environment/context constraints, such as whether they are outdoors, in a waiting room, with a companion or in a time-limited situation. On a larger scale viewing habits and preferences might depend on geographical location, or the availability of local content.



The objective is to analyse IPTV consumption – linear and non linear TV, Video on Demand and streaming services such as Netflix – in order to adapt value added service propositions to the viewers’ preferred convenience and schedule. This will involve regular IPTV viewing (set-top-box based, where a set-top-box is a device that turns different input sources into content that can be displayed on a TV or other display device), as well as mobile app based (MeoGo), which will require identifying viewing contexts, to best suit our offer both to the viewer’s interests and for specific usage contexts. Viewer behaviour becomes particularly important given the increase in non-traditional consumption via the web or mobile. This requires that the online IPTV offering must become adaptive and sensitive to fine tuning.



Activity Logs: These logs include the large majority of activities that are performed by IPTV viewers,  including both TV usage and viewing pattern information. Records of activity logs represent activities executed on each IPTV device (set top box or MeoGo mobile app) by one or more viewers.

Electronic Programme Guides (EPGs): This is a data stream comprising all the information about programmes and channels, such as scheduled date and time, description, duration, etc. To enable more agile information sharing, the EPG divides Programme and Channel into different input files.

Programmes catalogue: Contextual data providing additional insight on aired programs, compiling both external and internal knowledge, which complements the EPG – e.g. with programme genre, director and cast.

Commercial catalogue : Description of IPTV commercial offer, detailing the content of its different services – linear and non-linear TV packages, Video on Demand, Netflix and value added services, including associated costs and available bundling alternatives.

More detailed information about the data can be found in our data catalogue


Expected outcomes

  • Machine learning based models enabling detailed profiling of the IPTV client base, including information on the most relevant criteria associated to each profile;
  • Profiling models derived from the above
  • Evaluation of different profiling implementations and identification of the most adequate to the different business problems;
  • Trainable models allowing recommendation of IPTV content – e.g. TV programmes – and services – e.g. bundles and value added services;
  • Identification of viewing contexts, to adapt the IPTV content offer both to the client’s interests and to its specific usage context
  • Insights the customer service teams can use to refine existing bundles/price, adapting them to specific profiles.

Expected impacts

  • Reduction (15%) of Customer Rejection – This KPI measures the rate of rejected recommendations provided to viewers. This is measured by the number of times each customer explicitly rejects a particular recommendation (content or service offer).
  • Increase (15%) in Customer Engagement – This KPI measures the viewer engagement with recommendations. This is measured based on the uptake of recommendations (content or service offer) for each viewer.
  • Increase (5%) in Average Revenue Per User – Based on viewer uptake of value added content, new service offers or migration to bundles with increased margin.
  • It is the aim that an additional impact, relevant given the multiplicity of business interests of the Altice Group (telecommunications, media, advertising, financial services), will be the ability to extend the profiling and recommendation processes across universe of services provided by the Group. Cross-service profiling might allow tapping non-evident cross-selling opportunities and the subsequent impact on average revenue per user (ARPU) and retention figures.