Do it AI! It’s deep learning! Even if you say that, there are still many people who don’t come to the point, right? Me too.
I know, I know, but would you like to make a proposal at next month’s planning meeting or brainstorming? I mean …? Well, honestly, I don’t think there are many. It’s a little too future. It’s too far. ..
So, this time, we will introduce the technology of that area to the service as soon as possible ⇒ We will introduce about 3 carefully selected cases that have achieved great results in terms of cost cutting and user experience.
It is troublesome to read the commentary one by one! I will send it in slide & summary style for busy people.
Hot Pepper Beauty x Image Analysis AI mechanism and aim, understanding the difficult points of the actual project
Rough content introduction
- Adopted CNN to analyze the user’s “sense” and improve the service. Explanation of the flow and mechanism (super easy to understand)
- Introducing a case study using CNN to post “similar nails” in nail search in Hot Pepper Beauty
- There is no learning data! Discrimination accuracy does not improve! There is noise in the similarity calculation!Explanation of problems and real solutions that hit the real world
- Owned media: Introduction of inappropriate image NG processing cases and their processes in gathering
Since the creator of the slide is a professional with big data, it’s like “I’ve improved this much with AI!”It’s not a top story, it’s already a messy storyIs introduced.
I understand the theory. So, what kind of problems would occur if we actually did it, and how many people would do their best to clear them? A very rare slide that explains AI introduction cases from the perspective of.
The overall structure is rather long, but the reading response is perfect.
Of nail imagesDiscrimination accuracy 18%From, while internally creating gorigori learning data (a male team who does not know nails) and searching for parameters, mass-producing unidentified mystery list dataLiterally “do something”The story up to is a must-see.
The technical explanation part about image analysis is also summarized in a simple and easy-to-understand manner.
Image analysis / classification case study slide for Japan’s largest fashion app “iQon”
I tried to classify product categories using deep learning
Rough content introduction
- Limitations on classifying apparel products by name, tag, and description
- How to choose a deep learning model and judge algorithm
- A smart introduction to how to create and load efficient learning data
- Introducing how to incorporate it into the crawler
- It even publishes the actual server configuration (although it is simple).
Actually encountered in apparel services“You can’t automatically categorize products just from text information!”This is a story about how to tackle such a real problem with deep learning and solve it brilliantly.
Perhaps it was made for “unfamiliar people” like me, and I’m grateful for the structure that can be taught step by step in a very easy-to-understand manner.
From the former human-powered categorization, text discrimination and automatic sorting by morphological analysis, which became popular soon. And let’s let AI do the sorting work that would have been difficult and laborious. ThatI’m not sure what to start withReally clear about“You can do it in this order from here.”Let me understand.
I think the business impact is cost reduction and UX improvement, but the effect was probably not odd.Worth a lookis.
Story summary slide that explored user voice with artificial intelligence (others)
Rough content introduction
- Introducing design thinking and promotion of service development that creates “collaboration” from marketing research
- Colloquialized user voice x thousands of dying stories
- Opinion resolution is blurred in the questionnaire ⇒ Explanation of how to collect user’s voice and subsequent analysis
- Explains the design of supervised AI and automatic discrimination by this
This is Recruit’s slide again, but this is an example of its use in community services for young people such as “Seriously Club”. To tell you the truth, there isn’t much talk about AI and deep learning.
But in the actual fieldItems that would not normally be recognized as “issues”(Well, it can’t be helped … the part that makes you think like that), is this still? Thinking that, we examined measures and methods including automation by deep learning, and the story until it leads to the actual result is extremely hot.
It is a level that makes you want to imitate it.
After all, any advanced technology has “people and money”.
Well, how was it? Even if you haven’t seen all the slides yet, I’ve read all the slides! Did you feel the real AI x business initiatives that have already begun to move?
No matter how convenient it is, there is no end to itNeed preparation and sittingAnd thereFunds to pourIt doesn’t work without it.
In that case, nothing will start and will not proceed unless there is a person who draws money by highlighting the problem.
So, when you want to take on that role, that party can talk“Only what the person can imagine in his brain”I think it is. Most of the time.
It’s finally GW, and it may be good to read it carefully while writing a memo or a mind map.