Leah Moore, one of our brand journalists in the marketing department, sat down with in-house AI expert Cathy Feng to discuss recommendation engines, including our own RFP platform.
Leah: How do we keep our AI recommendation engines on track?
Cathy: There are two parts. From a user perspective, data is created when users click any items on the platform, stay on the page for a certain period of time, or like or dislike content. Actions and behaviors like these are fed back into the system, and help the recommendation engine to learn and improve.
On the other hand, in a business domain, user behavior and the document or content itself may not provide the best results. There could be bias. Having an architecture in place to bring background or domain knowledge to bear will be helpful here. In business situations, expert knowledge contributes to making the engine perform better.
We also have mechanisms and configurable parameters that allow us to check the model’s performance regularly and improve the recommendation results. As a result, the AI engine will be continually improving over time.
Leah: Can you explain AI to someone who may not fully understand it?
Just think about a human. Humans have eyes so that we can see and do things like read text. We can watch videos and see pictures. Humans have ears so that we can hear sound. We have mouths so that we can speak. In all those biological communications processes, we are taking raw data points in, processing them using our brain, analyzing them for useful information, and determining actions.
Instead of using the human brain, we can now use machine learning and deep-learning techniques to process more raw materials, get insights, and then take action. We call this AI, or artificial intelligence.
In cases where you are processing text with AI, we call this technology natural language processing or NLP. If you are processing videos or images, that’s computer vision, image recognition. If you process sound and voice, that’s audio processing. And on and on.
Leah: In that vein, what is a recommendation engine?
Cathy: Recommendation engines are AI tools based on user behavior, sometimes they also refer to user profiles to recommend the most relevant item to the customer.
Leah: Can you expand on what a user profile is?
Cathy: So, in the B2B domain, when we talk about user profiles, they can be more specific. For example, we have certain clients using our digital platforms. The user profile we are referring to could be the designation, the department, the business unit of the user, as well as the person’s role.
There are certain kinds of things of most interest to them. Of course, the B2C domain is more dynamic. A lot more content can be considered part of the user profile.
Leah: You already discussed our RFP recommendation engine. What are other ways can a recommendation engine can be useful to a business?
Cathy: So, in this platform, we recommend the most relevant responses to the user. We have also applied the recommendation engine to help our solution architect decide which services or solutions could be of interest to our customers. It’s already integrated into our own digital platform and then given to our people for use. And the engine could also be a good plus for knowledge-management portals for any organization, helping them make the most out of their data assets.
Leah: When you look into the future, how do you see this technology continuing to evolve? Do you see recommendation engines continuing to need human oversight or operating more independently?
Cathy: This is one area that has good adoption in the market. It’s still an evolving solution in the B2C domain, and we are also seeing the trend in the B2B space. More and more applications are starting to adopt a recommendation engine. Maybe one day it will evolve from something good to have to something businesses must have.
Of course, there are challenges involving improving performance. Many things need to be done to improve recommendation engines, such as combatting deep bias, optimizing the computational cost, personalization, et cetera, et cetera. Resolving those matters will be crucial for these applications to take off in the B2B world.
Regarding human oversight, if you refer to the feedback loop, yes. Human intervention or human involvement will always continue. Humans need to be there, as a recommendation system is a two-way system. If you are talking about expert knowledge, it is good to incorporate the knowledge into the system.
Finally, to answer your latter question, yes, the operation will be more independent. The AI engine will operate more independently. That is already happening as it becomes more sophisticated, and the algorithms mature.
