Top Page | Upper Page | Contents | About This Site | JAPANESE

Recommendation System

The recommendation system has become an everyday technique for making recommendations on the Internet . It may be displayed as "Recommended for you", or even if it is not written that way, what is displayed may be in the order of priority.

How the recommendation works

The recommendation system has already become a daily technique, so let's try to unravel this technique in reverse engineering.

First of all, in the recommendation system, it is recommended by a mechanism such as "People who are similar to you choose this" and "People who choose this also choose this".

So, if you have the following information in the system, you should be able to create this mechanism.

Data Science techniques that show this can include Associations Analysis , Cluster Analysis , and Matrix Decomposition .

Special for machine learning

The technique used in the recommendation system is a type of " Machine Learning ", but the recommendation technique does not appear much in the general explanation of machine learning. There are several possible reasons why it does not appear, but one of them is that it is difficult to classify, as shown below.

In general, when classifying into supervised learning, unsupervised learning, reinforcement learning are used for prediction and reasoning. Unsupervised learning is often described as a technique used in data preprocessing and data comprehension.

On the other hand, recommendations are unsupervised learning techniques, while recommendations are predictions and reasoning. Recommendations are special in this respect.

By the way, there is a One-Class Model as a way to use unsupervised learning for prediction and reasoning, but the recommendation system is quite different.

Application of recommendation system to other fields

In data science and AI, we often do PoC (proof of concept) and do not go any further. On the other hand, the recommendation system is becoming more and more advanced as it becomes a daily technology. I think this is the number one successful example of a successful combination of data science and business models. By the way, is No. 2 image recognition and No. 3 voice recognition? .. ..

The recommendation system may have applications as a technology for making decisions from a wide variety of combinations.

In addition, many of the cases where PoC stops are based on the technology, and it seems that confirmation of compatibility with the target is secondary, so I think that it is a good subject to think about such things. increase.

Without being bound by classifications such as supervised learning and unsupervised learning, I think that it may be a good subject to consider the relationship between the algorithm of the method and the relationship between input and output and the target business or business. increase.

Next Business Model