How Colombia saw a 3x jump in contextual recommendations with smarter NLP

Share this post:

Mobile has changed everything. Thanks to phenomenal smartphone penetration and democratization of data, more and more consumers are getting online and consuming content like never before. Indeed, forecasts predict India’s internet users will number 829 million by 2021.¹

What’s more interesting to note is the rapid growth of Indian language internet users. By 2021, the number of Indian language internet users is expected to touch a whopping 536 million. In fact, regional language internet users are expected to account for nearly 75% of India’s internet user base by 2021.²

This has led businesses to think: is there a smarter way to take advantage of this massive market potential? Be it improving customer experience by real-time personalization or updating campaigns based on changing customer response or simply building better relationships to improve conversion, data and cognitive insights can truly take marketing campaigns to the next level.

One such innovation was carried out by Colombia, one of the largest publisher-owned ad network platforms in APAC. Through its premium content network, the platform connects brands to over 270 million users. Building on the fact that over the next five years 9 out of 10 new internet users are going to be regional language consumers,² the company was looking to tap into the vernacular content segment and grow its user base.

Colombia was looking for a technology partner who could help address the growing need for varied content formats for targeting their publishers as well as readers. Regional language content being the key, the ability to promote content across language was a critical necessity to stay ahead in a competitive environment.

The challenge of promoting vernacular content

Colombia was looking for a solution that could help them promote regional content by seamlessly implementing an accurate cross-language recommendation system over multiple media, which included text articles, audio, and video. The company tried multiple solutions but did not see any significant improvement as it could address only part of the problem.

NLU and accurate targeting

Initially, Colombia built a holistic recommendation engine with more than 25 different algorithms competing with each other to churn out the most relevant results in real-time to give their end-customers what they seek. It then tried data enrichment with some open-source software and standalone solutions, but the results didn’t rise to expectations.

Then it found that IBM Watson’s Natural Language Understanding (NLU) features best performed the tasks it sought to do.

Watson NLU service

The Watson NLU service analyzes keywords, concepts, and categories in the articles. With Entities, one can extract the people, places, and organizations mentioned in the text. It helps publishers understand sentiments (positive or negative) that readers express, including the ones expressed towards a particular keyword, entity, or phrase.

Finally, publishers can develop custom models using IBM Knowledge Studio and use them with Watson NLU to tailor articles for specific audiences using domain-specific entities and relations. Watson helped Colombia enrich their content with named entities, semantics, and sentiments that significantly improved their upstream, real-time recommendation algorithms.

Watson’s NLU feature also plays a crucial part in low-level data enrichment. This is further fed to upstream deep learning and traditional machine learning models, enhancing their accuracy – in classification and recommendation models.

What’s more, IBM Watson could integrate with the recommendation engine that Colombia had earlier built.

Steep rise in contextual recommendations

After deploying IBM Watson’s NLU suite, Colombia surpassed its earlier Natural Language Processing (NLP) benchmarks based on user interaction, engagement, and retention as more and more people started clicking on, reading, and viewing the content they recommend. The company reported a three-fold increase in contextual recommendations with a focus on cross lingual endorsements.

IBM Watson played a critical role in building a tech stack for Colombia to help strengthen their publishers’ digital presence beyond regular social channels. Additionally, their readers’ interaction with the platform in terms of retention, viewership, and readership significantly increased. Colombia sees IBM Watson as a strategic partner in the field of deep learning as it charts its way forward in regional content synthesis and targeted information dissemination.

==========================================

References

1. https://gadgets.ndtv.com/internet/news/indias-internet-users-to-double-to-829-million-by-2021-says-cisco-1710120
2. Indian languages – Defining India’s Internet: A study by KPMG and Google (https://assets.kpmg.com/content/dam/kpmg/in/pdf/2017/04/Indian-languages-Defining-Indias-Internet.pdf)