Hike, Artificial Intelligence and Machine Learning
AI and ML can be used for effective and seamless exchange of communication and expression among close friends.
Homegrown internet startup Hike describes that social products should be joyful. They should be built around people and not the other way around. They should be fun and should celebrate the depth of relationships. They should allow people to be their true selves and go beyond the limits that hold them back in the real world. And following the same, their messenger app is built around the same values and policies.
Hike Sticker Chat has integrated AI and ML in the app and to learn more about it, we spoke with Dr. Ankur Narang, Vice President - AI and Data Technologies at Hike.
How AI and ML works in suggesting stickers on Hike Sticker Chat?
Dr. Ankur: Hike Sticker Chat is on a mission to reduce people’s dependency on the keyboard. Unlike other countries such as US & China, dialects in India change every few kilometers.
This has a direct impact on input, where users struggle to communicate in their local dialects as QWERTY keyboards aren’t able to keep up. This is the gap we’re looking at bridging. Through stickers we are aiming at two key things, creating Sticker Content to cover a large chunk of local vocabulary & using ML & AI to scale this up drastically. Secondly, Sticker Discovery, recommending the right sticker at the right time, which sits at an interesting intersection of ML & AI taking into account a dynamic range of factors.
We believe AI and ML can be used for effective and seamless exchange of communication and expression among close friends. Keeping that in mind, we are continuously driving locally relevant innovations and working on some unique elements in NLP to develop unique solutions that can make sense of text in local languages written in the English script. Case in point, our research will also help us create better customer experiences backed on seamless AI &ML integration. We continuously aim to solve the input problem by suggesting sticker that is backed by AI & ML, infinite content creation and voice input which helps in suggesting the right stickers at the right time in different local languages and dialects making day to day conversation seamless and expressive.
Focus on Research and Development
Dr. Ankur: Hike truly believes in a culture of research that is led by innovation. We’re working closely with the ecosystem to enable the field of AI & ML. We believe that multiple collaborations across the ecosystem can best leverage the bright minds of our country. Some of our initiatives so far include, academic partnerships with entities like IIITD and strong focus on research through papers. We showcased our latest work at the prestigious IJCAI platform where we presented a paper on the role of AI integration in Hike Sticker Chat for sticker recommendation. Prior to that we published a paper on Friend Recommendation earlier at ECIR 2019 where out of 151 submissions only 44 (29 per cent approx.) short papers were accepted and Hike’s ‘Heterogeneous Edge Embedding for Friend Recommendation’ was one of them. Some of Hike’s other initiative is to encourage the AI & ML ecosystem in India that includes— one of the biggest ML hackathons in the country, white Papers & internship opportunities for students.
Today, Hike is the only player in the market to be solving problems at such massive regional volume with a focused inclusive and hyper-local approach catering to low-end handset & low bandwidth market areas.
At Hike, how differently are you using emerging technologies like AI and ML?
Dr. Ankur: Hike Sticker Chat is on a mission to reduce people’s dependency on the keyboard and solve for local markets. Unlike any other country, languages & dialects in India change every few kilometers. This creates a direct impact on input, where users struggle to converse in their local language/dialect as QWERTY keyboards aren’t able to support that. We are aiming to bridge this gap by using stickers as a foundation. Through stickers we are aiming at two key things:
- Creating Sticker Content to cover a large chunk of local vocabulary and using ML & AI to scale this up drastically
- Secondly, Sticker Discovery, recommending the right sticker at the right time, which is backed by AI/ML capabilities
For example, while we do have about 60,000 stickers, for us to cover a large chunk of local languages and dialects, we are looking at aggressively scaling this base even further and reaching 100,000 stickers before the end of the year itself. This is where we see AI & ML at the heart of the product, enabling scale and making the overall experience for the user super seamless & qualitative. We are the only company that is building big on AI/ML and we are heavily investing in areas such as Natural Language Processing (NLP), Computer Vision and Social Network Analysis (SNA).
What are your future plans?
Dr. Ankur: Earlier this year, we decided to unbundle Hike the super app into multi-apps and launched Hike Sticker Chat in April 2019. After four months of the launch, Hike Sticker Chat has more than 1 million Weekly Active Users spending more than 231 mins per week on the app. There are more than 50 million stickers being exchanged weekly and more than 60,000 stickers available for our users to chat with and express better through them. Our ultimate mission is to reduce people’s dependency on the keyboard- type less and express more with stickers and solve the input problem for local markets. This can only be accomplished by using the right mix of AI and ML in sticker recommendations and our present focus lies on building a world-class AI & ML team that helps in enabling a culture of research-led innovation. In the coming months, you will see some amazing developments at hike that are going to be really exciting for the users making their experience more seamless and will definitely scale up the product further.
Hike is creating stickers for local markets in different local languages. How are you leveraging technologies like NLP and AI in order to facilitate this?
Dr. Ankur: Like we said, the users are looking for a seamless chatting experience where input and language isn’t a hindrance in conversing with their close friends. The hindrance arises when there is no local language/ dialect input support because chatting amongst close friends mostly happens in local languages and dialects. However, due to the lack of localized input options, people type in the English. There are keyboards that help transliterate input from the QWERTY keyboard to local scripts though none of these options make the user feel as productive as when they are texting in English. For a country like India that has heterogeneous languages and dialects, the problem becomes more severe. At Hike, we believe AI and ML can be used to find an effective solution to this problem. We are currently working on some unique elements in NLP to develop unique solutions that can make sense of text in Indian languages written in the English script along with deep learning, Social Network Analysis (SNA) and Computer Vision. We are also exploring how voice input can be used in conjunction with visual expressions to define the next generation of communication in the virtual world.
How big is the AI team at Hike?
Dr. Ankur: The AI/ML team at Hike is the most dynamic team that is behind all the innovative work that is enabling hike’s vision of solving for local markets. The team is a diverse mix of young talent and experienced names in the ecosystem working closely together, making this one of the best examples of a lean team working on long term scalable solutions. Earlier this year, Hike has also launched a programme called ‘ZeroTo2’ to hire young engineers with 0-2 years of experience. Hike’s user base typically consisted of users in the 18-24 years age bracket but now we have decided to widen our focus to include even younger users starting with 16-year-olds. We are undertaking an extensive training to hire the right talent for the organization. We created this team keeping in mind some of our principles of focusing on young talent, building a lean and highly productive team and most importantly creating a symbiotic stream of learning within the company and within the ecosystem. This team consists of 15+ members specifically working on the areas of NLP, Computer Vision & Speech. We are currently looking at expanding this team and are running a range of specialized hiring drives.