Blog

<h2>Overview</h2>
Earlier this year, the CRE ML Team's research paper on violation detection systems in chat messages was accepted to the <a href="https://sites.google.com/view/ecnlp">ECNLP</a> (European Community for 
Neuro-Linguistic Programming), a workshop of the ACL (The Association for Computational Linguistics).
<a href="https://aclanthology.org/2022.ecnlp-1.8/">Textual Content Moderation in C2C Marketplace - ACL Anthology</a>
The <a href="https://aclanthology.org/">ACL</a> is one of the industry’s most prestigious language processing conferences. ACL 2022, the 60th workshop of its kind, was held in Dublin, Ireland.
<h2>Key points of presentation</h2>
<ul>
<li>Implemented a machine learning system to detect violations of the terms of service written in chat messages</li>
<li>Improved violation detection accuracy while significantly reducing the number of manual checks by customer support</li>
<li>Created a system that is easy to maintain and continuously improves detection accuracy</li>
</ul>
<h2>Background</h2>
After an item is purchased on the Mercari marketplace, the seller and buyer can freely communicate with each other via a chat system. To facilitate smooth transactions, harassing behavior such as making defamatory or threatening comments, is prohibited in Mercari’s terms of service. However, it is difficult for customer support to visually check all such communications, and therefore Mercari uses a system that automatically screens for violations.
In the past, the system used heuristics such as keyword searches to detect violations, but the detection accuracy was limited. However, because trouble can easily follow communications between sellers and buyers, and there is a significant risk of overlooking a violation, it is necessary to ensure the reproducibility of violation detection (also known as recall). An additional issue was that the number of false positives remained high, even if a conversation had passed the heuristic screening, and therefore customer support representatives had to check huge volumes of conversations manually.
In this study, researchers implemented a system that used a machine learning model specialized for the aforementioned screening process. 
The study reported that implementing the system reduced the number of false positives while ensuring reproducibility, resulting in a significant reduction in the number of conversations that needed to be checked manually.
<h2>Summary of paper</h2>
After a buyer purchases an item on the Mercari marketplace, the seller and buyer are able to freely send chat messages to each other. However, there are cases where the content of these messages makes sense in the context of the parties’ conversation that started on the separate public chat message board used prior to the item purchase. We therefore devised a machine learning model that makes inferences based on the context of the previous conversation, rather than estimating whether or not the text in isolation is in violation of Mercari’s terms of service.
The model itself is designed to be extremely lightweight, and can make inferences at a cost that is reasonable in relation to the vast number of chat messages exchanged on Mercari. In addition, the architecture allows each type of violation to be added, modified, or deleted with ease since the content of violations stipulated in the terms of service may be modified in the future to adapt to changing social norms. The moderation staff then checks the conversations that the machine learning system has screened. The decisions made at this time are added to the training data as true labels. In this way, a feedback loop is created within the violation detection system, which is expected to improve detection accuracy continuously.
In addition, Mercari confirms chat messages from the viewpoint of preventing fraud and trouble, in compliance with laws and regulations, and with the consent of each user.
<h2>About CRE ML Team</h2>
The CRE ML team operates several other machine learning services and provides technical support for the safety and security of our users.

Mercari AI’s research paper—”Textual Content Moderation in C2C Marketplace”—accepted to ACL2022 ECNLP Workshop

Hi! I&#8217;m <a href="https://twitter.com/MLBear2">ML_Bear</a>, an ML Engineer on Mercari&#8217;s Recommendation Team. 
I&#8217;m in charge of improving the recommendation components displayed on the Mercari home page and elsewhere. Today I&#8217;m going to show you examples of how I improved the recommendation logic by using Mercari&#8217;s massive amount of log data and the metadata assigned to items sold on the Mercari app by users. (More details on the metadata below.)
<h2>Item Metadata</h2>
To improve the item search experience for users, Mercari assigns various kinds of metadata to the items on the app. For example, metadata for apparel includes color and the fabric&#8217;s texture, while for home appliances it contains model numbers. The different kinds of data that indicate an item&#8217;s traits is what we call metadata. 
I&#8217;d like to take a look right now at the metadata we assign to books and manga (what I&#8217;ll refer to as “title data” below). 
In the Mercari app, metadata is assigned to books and manga to show what series each item belongs to. For example, &quot;Kingdom Vol. 1,&quot; &quot;Kingdom Vol. 2,&quot; &quot;Kingdom Vol. 25-50,&quot; and &quot;Complete set of Kingdom&quot; get the same &quot;Kingdom&quot; title data. 
At Mercari, we also record all logs of items viewed by users. 
So, I tested a way I thought we could improve book and manga recommendations to match user interests by reconciling those logs and title data while using the Item2Vec recommendation technique.
<h2>Using Item2Vec</h2>
Word2Vec1 is one well-known technique for natural language processing.It is a method that proposes words can be processed in a variety of ways by vectorizing them. Although it&#8217;s a fairly old technology, it is highly versatile, so it&#8217;s used in many kinds of research and in machine learning competitions. 
Item2Vec2 is a way to apply that technique to item recommendations. 
Implementation and training are relatively simple, and it&#8217;s a very easy way to produce accuracy. That&#8217;s why other companies have also shared use cases. It&#8217;s difficult to use if there aren&#8217;t many user logs, but personally, I think it&#8217;s highly suited to a service like Mercari that has a massive number of users and an enormous number of logs collected on a daily basis. 
I created a recommendation logic with the following method.
<ol>
<li>
Derive the similarity and relationship of each item. (With Python, process monthly by using data from N past months.) 
a. Extract users who viewed book or manga items over the past N months. 
b. Collect users&#8217; item browsing histories over the past N months. 
c. Extract arrays of titles viewed by session. 
d. Regard title arrays as sentences and each title as a word, then derive title vectors with Item2Vec. 
e. Use title vectors to derive relationships between titles (cos similarity). 
f. Store data on relationship between titles in BigQuery.
</li>
<li>
Generate recommendations for users. (With BigQuery, process hourly by using data from N past hours.) 
a. Extract the history of items browsed by the user over the past N hours that are also tied to title data. 
b. Derive &quot;titles similar to item titles the user has viewed.&quot; 
c. Store the data derived in the preceding step as &quot;list of titles to recommend to users&quot; in Bigtable. 
d. If there is a title list to recommend in Bigtable when a user visits the home page, search for those titles and display them in the recommended item list.
</li>
</ol>
<img src="https://storage.googleapis.com/prd-engineering-asset/2022/09/992a7635-image2.png" alt="" /><center> 
Fig1. Concept diagram for separating titles by sessions in item browsing logs
<img src="https://storage.googleapis.com/prd-engineering-asset/2022/09/3d14bef2-img_1-473x1024.png" alt="" /> 
Fig2. The new recommendation section&quot;Recommended Books &amp; Manga for You&quot; 
</center>
I won&#8217;t discuss details about the system when it generates recommendation data or displays it in the app, but below is an overview (everything is processed on GCP)
<ul>
<li>Process most data concerning item viewing logs with BigQuery (BigQuery makes it easy to process hundreds of millions of lines of log data.)</li>
<li>Use Gensim on Colab to process Item2Vec calculations.</li>
<li>For the data pipeline, use Cloud Compower (Airflow) to execute hourly and monthly data processing.</li>
<li>Use Bigtable for the data store in the production environment.</li>
</ul>
<h2>Item2Vec Results</h2>
I&#8217;d like to show you some of the results from calculating with Item2Vec. 
Below are the results from the most_similar function in the model I calculated with <a href="https://radimrehurek.com/gensim/">Gensim</a>. (The number indicates the degree of similarity.)
Titles similar to the manga &quot;Dragon Zakura&quot;
<pre><code>1[(&#039;Dragon Zakura 2 17&#039;, 0.8193),
2 (&#039;Angel Bank: Dragon Zakura Gaiden&#039;, 0.6674),
3 (&#039;Let&#039;s Go to University of Tokyo: Dragon Zakura Official Guidebook&#039;, 0.628),
4 (&#039;7 Skills from Dragon Zakura for Parents to Bring Out the Ability for Kids to Pass the University of Tokyo Entrance Exam: Longshan High School “Parent Skills” Special...&#039;, 0.5366),
5 (&#039;Dragon Zakura Official Supplementary Reader, Age 16 Textbook 2, Where Do Study and Work Connect?&#039;, 0.5265),
6 (Gin no Anchor: Naitei Ukeoi Manga&#039;, 0.5145),
7 (&#039;The Great War of Archimedes&#039;, 0.5118),
8 (&#039;Bucho Shima Kosaku: Bilingual Edition&#039;, 0.5098),
9 (&#039;Investor Z Complete&#039;, 0.5034),
10 (&#039; Dragon Zakura Official Supplementary Reader, Age 16 Textbook, Why Study? What to Study?&#039;, 0.4913),
11...</code></pre>
Dragon Zakura is a Japanese manga series written and illustrated by Norifusa Mita. 
It tells the story of high school students studying to enter the University of Tokyo. 
Supplementary readers for entrance exams are mixed in with Norifusa Mita&#8217;s works. I found this unexpected result very interesting.
Titles similar to the manga &quot;Shoujiki Fudousan&quot;
<pre><code>1[(&#039;24 Pieces of Divine Knowledge for Beating Real Estate Companies&#039;, 0.5547),
2 (&#039;Stand Out at All Stages, From Novice to Veteran: The Most Powerful Real Estate Investing Textbook for Investors 1...&#039;, 0.493),
3 (&#039;Kurosagi&#039;, 0.4731),
4 (&#039;The Easiest Real Estate Investing Textbook in the World, 1st-Year Students: Perfect for Beginning Again&#039;, 0.4692),
5 (&#039;Rumored Secret Drugs in Industry: Big Real Estate Techniques&#039;, 0.4675),
6 (&#039;Souzoku Tantei&#039;, 0.4628),
7 (&#039;The Future of Real Estate: Prepare for the Big Housing Turnover&#039;, 0.4405),
8 (&#039;Worst Properties of the Year&#039;, 0.4351),
9 (&#039;Real Estate Investing Don&#039;ts&#039;, 0.4313),
10 (&#039;Monkey Peak&#039;, 0.4291),
11...</code></pre>
This is the story of a real estate agent who can&#8217;t tell a lie. The list includes &quot;Kurosagi,&quot; which Takeshi Natsuhara (the original writer of &quot;Shoujiki Fudousan&quot;) worked on. There is indeed a strong connection because the original writer is the same. It&#8217;s also interesting that the list includes regular real estate books among the manga. 
By the way, I only just now learned about the manga &quot;Souzoku Tantei&quot; from writing this tech blog, so I took a break from writing, bought the digital edition, and gave it a read. It&#8217;s pretty good. It certainly feels similar to &quot;Shoujiki Fudousan.&quot;
Titles similar to the novel &quot;Clouds Above the Hill&quot; by Ryotaro Shiba
<pre><code class="language-1[(&#039;Tobu">2 (&#039;Yo ni Sumu Hibi Complete&#039;, 0.7067),
3 (&#039;Natsukusa no Fu Vol. 1 &amp; 2&#039;, 0.6681),
4 (&#039;Owl&#039;s Castle&#039;, 0.6624),
5 (&#039;When the Last Sword Is Drawn Vol. 1 &amp; 2&#039;, 0.6594),
6 (&#039;Kashin Vol. 1-3&#039;, 0.653),
7 (&#039;Sekigahara Vol. 1 &amp; 2&#039;, 0.6471),
8 (&#039;Oda Nobunaga&#039;, 0.647),
9 (&#039;Komyo ga Tsuji&#039;, 0.6468),
10 (&#039;The Last Shogun: The Life of Tokugawa Yoshinobu&#039;, 0.6324),
11...</code></pre>
There are lots of works by Ryotaro Shiba.
Titles similar to the novel &quot;Downtown Rocket&quot; by Jun Ikeido
<pre><code>1[(&#039;Downtown Rocket: Ghost Yata Glass&#039;, 0.9246),
2 (&#039;Downtown Rocket: Yata Glass Ghost&#039;, 0.9115),
3 (&#039;Stock Price Crash&#039;, 0.8829),
4 (&#039;Bank Manager Special Assignment, New Edition&#039;, 0.8819),
5 (&#039;Hanasaki Mai Speaks Out&#039;, 0.8758),
6 (&#039;Bitter Enemy&#039;, 0.8756),
7 (&#039;Scandal, New Edition&#039;, 0.8702),
8 (&#039;Bones of Steel&#039;, 0.8686),
9 (&#039;Financial Detective&#039;, 0.8614),
10 (&#039;Ginko Shiokinin&#039;, 0.8589),
11...</code></pre>
This list also has many works by the author, in this case Jun Ikeido. Mercari users who search for novels might view many different works by the same writer.
<h2>Recommendation Logic A/B Testing</h2>
After completing Item2Vec, I ran A/B testing (also cleverly named: Book2Vec). 
At Mercari, it&#8217;s a big part of our culture to quantitatively assess improvements with A/B testing. Team member yaginuuun released materials about our attempts to standardize A/B testing, so check those out. (<a href="https://speakerdeck.com/shyaginuma/btesutobiao-zhun-hua-hefalsequ-rizu-mi">Slides</a> * This slide is in Japanese only) 
In conclusion, it produced good results (though I can&#8217;t share the exact numbers) and the recommendation logic I made was adopted. I&#8217;m diligently developing it so pages in addition to the home page can use it too. 
The joy I received from having the logic I created and implemented run on a service used by tens of millions of people is the best part about working as a Mercari engineer.
<h2>We’re Hiring!</h2>
In addition to metadata for books and manga, we intend to improve the recommendation logic we use for the massive amount of other data at Mercari. 
As regular users of the Mercari app may be aware, the Mercari recommendation logic still has lots we can do with it and plenty of room for improvement. 
The fact is, because of the massive number of visitors our site gets, a complex logic poses challenges because we can&#8217;t use it in the actual environment. On the other hand, however, if we have just one that works, then it will have extensive economic value. I want to keep on working out ways for users to find the items that are right for them. 
If you&#8217;re interested in this work, I&#8217;d be happy to hear from you, as would my Mercari colleagues, so feel free to contact us. See ya!
<a href="https://apply.workable.com/mercari/j/3333DC5597/">Software Engineer, Machine Learning &amp; Search &#8211; Mercari</a>
<h2>References</h2>
[1] Mikolov T, Sutskever I, Chen K, Corrado GS, Dean J. Distributed representations of words and phrases and their compositionality. In Proceedings of NIPS 2013 (pp. 3111-3119). 
[2] Oren Barkan and Noam Koenigstein. 2016. Item2vec: Neural item embedding for collaborative filtering. arXiv preprint arXiv:1603.04259.

Attempt to improve item recommendation accuracy using Item2vec

The MERCARI GEARS Lecture Series is a video series for engineers that features the future direction and technical challenges of each Mercari Group company, including Mercari Inc.
In this edition of MERCARI GEARS, Mercari's Edge AI Team highlights the Mercari app’s real-time item scanning feature, available on iOS and Web.
<center> 
<iframe loading="lazy" width="560" height="315" src="https://www.youtube.com/embed/cKBzwB5StLY" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> 
Edge AI - Realtime Item Scan for iOS and Web [AI@Mercari] 
</center>
Mercari has a feature called &quot;AI Listing&quot; that suggests category, brand, and even selling price for an item when you upload a photo of it.
To use the &quot;AI Listing&quot; feature, you used to have to take and upload a photo of an item. But now, the Edge AI Team has made it possible to to use the feature as soon as your device’s camera is activated and the item is in focus.
In order to create a quicker and more comfortable display experience, the team leveraged recent advances in smartphone technology to create an ML model with low computational cost. As a result, suggestions through this feature can now be made directly on the smartphone without having to go back and forth to a computer server.
We are working on a variety of things, including improving processing time and device compatibility to adapt quantization to the learning model and make inferences, developing tools to monitor performance after implementation, which are explained in detail in the video. Please take a look!

Edge AI  – Realtime Item-Scan for iOS and Web

“<a href="https://medium.com/mercari-engineering" title="Making Mercari">Making Mercari</a>&quot; is an engineering blog for Mercari's US product. AI-related posts on &quot;Making Mercari&quot; will also be featured on this website.
<hr />
Here is a post by Abhishek Munagekar, Software Engineer for the Machine Learning Platform at Mercari US. In this blog post, he describes how his team optimized the cost of computing resources used for machine learning training. More specifically, there were some challenges with downscaling of the Kubernetes cluster used for ML model training. Read the blog post to find out how PodDisruptionBudget, dedicated node pools and the Gatekeeper Assign CRD were used to overcome these challenges.
<a href="https://medium.com/mercari-engineering/how-we-optimized-our-machine-learning-training-infrastructure-costs-c2a64f7451bb" title="**How we optimized our Machine Learning Training Infrastructure Costs | by Abhishek Munagekar | Making Mercari | Apr, 2022 | Medium**">How we optimized our Machine Learning Training Infrastructure Costs | by Abhishek Munagekar | Making Mercari | Apr, 2022 | Medium</a>

How we optimized our Machine Learning Training Infrastructure Costs @ Mercari US

Interview

In <a href="https://ai.mercari.com/en/articles/ai/suadd_kimuras_1/">AI Listings, Image Search… CEO Shintaro Yamada and Director Shunya Kimura Look Back at Mercari’s AI Journey</a>, the two discussed the background and challenges that prompted Mercari to actively pursue the use of AI, which led to the release, starting in 2017, of a variety of features powered by AI technology, including AI listings, personalized timelines, and image search. In this article, the two discuss AI ethics and a concrete path forward to achieving Mercari's vision of a recycling-oriented society.
<h2>How Mercari arrived at &quot;recycling-oriented society” as its watchword</h2>
Shunya Kimura: When did Mercari start using &quot;recycling-oriented society&quot; as a watchword?
Shintaro Yamada: Probably starting around September 2019. Previously, we had a sense of the issue as an organization, but we started our internal discussion on ESG in earnest around January 2019. I felt that a recycling-oriented society was an issue that Mercari should confront head-on. First, we identified the materiality of Mercari to identify the issues that needed to be addressed, and then we published our first sustainability report.
Mercari currently has more than 20 million customers per month. In the future, we hope that our service will be used by 30 million people, 40 million people, and so on. To get there, I believe we should move beyond just &quot;convenient to have&quot; to become an indispensable part of society. In order to achieve this goal, we’ve had discussions with experts since last year to define our platform philosophy and the basic principles of our marketplace. I can now see a path forward for Mercari to fulfill its responsibility to contribute to a recycling-oriented society, which will also lead to our business growth.
<img src="https://storage.googleapis.com/prd-ai-asset/2022/02/4d2363c1-image1-1024x683.jpg" alt="" />
Over the last few years, interest in the natural environment has grown rapidly, partly due to the influence of the coronavirus pandemic. Global warming and massive generation of waste are also major problems. Our current wasteful use of resources is a problem that should be dealt with by society as a whole. I want to see a world where goods are recycled and used for as long as they are usable. We’re always thinking about how Mercari can become an indispensable part of such a recycling-oriented society. Thinking about how Mercari's services can deliver value to society today, I think “contributing to a recycling-oriented society” is a particularly important slogan.
<h2>What friction needs to be eliminated to close the 145 trillion yen gap?</h2>
Kimura: In terms of accelerating the reuse of goods, I think that Mercari is contributing to some extent. What do you think?
Yamada: We're still a long way off. Total gross sales in Japan, including non-reusable goods such as food, are valued at about 146 trillion yen per year in retail alone. The total transaction value of Mercari is 1 trillion yen, so we have not even reached 1%. As this difference shows, I believe there are still a lot of goods that are discarded or stashed without being listed on the second-hand market.
<img src="https://storage.googleapis.com/prd-ai-asset/2022/02/1b4220f5-image5-1024x683.jpg" alt="" />
Yamada: For example, if everything you have at home has been converted into data and you can complete your listing in an instant, it will be even easier to sell on Mercari. Right now, you still need to think a lot and put in a lot of effort to make a listing. There are still complicated aspects such as shipping methods. If we can remove that friction from the current situation, we’ll get more listings. Merpay makes the exchange of money even more seamless. I think we can make it a service that further circulates goods, where people think, &quot;I'm going to buy this, so I might as well try listing it.”
Right now, we have to actively buy and sell, but I think this is where AI can be very powerful. It can make suggestions like, “Since you’re buying this, why not put this up for sale?” If we can promote the effective use of resources through Mercari, we can accelerate progress towards a more recycling-oriented society.
<h2>The second-hand market alone is not the answer: partnering with retailers of new merchandise</h2>
Kimura: It’s quite interesting how AI can contribute to a recycling-oriented society. Mercari customers search for goods they want, but they don't actively look for goods they might potentially need, so Mercari uses AI to suggest such goods to them. Similarly, by showing the demand for goods the customer already has, you can create opportunities for listings. Mercari is CtoC but is expanding its domain to include BtoC with the launch of its online store service, Mercari Shops. It is now becoming a huge marketplace, more than just a flea market app between individuals. The cycling of goods will break down unless both the selling side and the buying side accelerate together. Matching up the numbers is also important in our business strategy. In order to contribute to a recycling-oriented society, we also need the cooperation of retailers of new merchandise.
<img src="https://storage.googleapis.com/prd-ai-asset/2022/02/569af45c-image3-1024x683.jpg" alt="" />
Yamada: I agree. To meet this challenge going forward, I believe it’s not enough to simply develop the second-hand market. We’re also working on an approach that communicates data on what’s tenable in the second-hand market to retailers of new merchandise. I believe that circulating data as well as goods can gradually move us closer to a world where manufacturers produce the products they need in the quantities they need and customers get as much use out of them as possible.
Currently, we’re gradually starting to establish data links with manufacturers. We’re not quite there yet, but we’re starting to see some areas where we can get some value. If you restrict yourself to the second-hand market only, the task of converting all the goods around you into data becomes much more difficult. Instead of trying to do everything in our own domain, we at Mercari need to connect more with retailers of new merchandise, including catalog (SKU) linking, while keeping data privacy in mind.
<h2>What are the AI ethics? Ethical issues at Mercari</h2>
Kimura: We already discussed the importance of AI as a technology for the future development of Mercari. However, I believe it’s equally important to discuss the topic of how to handle AI in an ethical manner. For example, with regard to the recommendation function, you shouldn't handle data in a way that violates customers’ privacy or make recommendations they find objectionable. Recommend what they need, but don't force unwanted information on them.
Yamada: Yes, it's important not to force unwanted information on them. Another thing I would like to be even more conscious of in the future is whether what we are doing will benefit humanity. Even for goods the customer needs, Mercari should have higher ethical standards because Mercari also handles the money part with Merpay. That’s why, for example, you shouldn’t be in a situation where you have to go into debt to purchase a product beyond your ability to repay. Mercari is a platform for buying and selling, so you might think that ethical issues would be less likely to arise than in advertising or social media. However, we want to avoid creating a system that makes people buy more and more for the sake of business growth. In that sense, AI development in Mercari is a place that offers an interesting experience. You’re likely to encounter ethical challenges that no one has ever thought of in the past.
<img src="https://storage.googleapis.com/prd-ai-asset/2022/02/f72b46c6-image2-1024x683.jpg" alt="" />
Kimura: Recently, when we were having an internal discussion about the ethics of AI, we talked about the possibility of sensitive issues that could cause discomfort and mistrust among customers when we predict products they may want to buy next. For example, they may have hobbies they don't want others to know about, or they may feel the recommendations interfere with their privacy.
Yamada: Yes, you need a balance between serendipity and recommendations, otherwise it can feel unpleasant. Because your activities on Mercari are not the only thing you do, it’s possible that you go camping on a daily basis, but don’t buy or sell camping goods on Mercari. We’re required to have a sense of ethics as to what kinds of products we recommend to which customers. That's the challenge we face going forward.
<h2>Data can reveal the intrinsic value: What's interesting about Mercari's AI and the path forward</h2>
Kimura: In order to achieve a recycling-oriented society, we’re working toward the goal of using AI to estimate supply and demand and deliver what is needed, in the amount needed. But to be honest, we still have a ways to go. We need to accelerate our AI modeling and personalization experiments, which we will continue to work on going forward.
Yamada: There's a lot to do, and there's a lot to try. We don't have as many AI/ML engineers as Big Tech. On the other hand, the data handled by Mercari includes a lot of information that is close to the customer, such as customer-generated product images, captions, and prices. Since there are few companies and services that have this kind of data, I believe there is potential for innovation.
Kimura: The type and nature of data is different from regular e-commerce. Understanding the intrinsic value of things is a very interesting part of the AI at Mercari.
<img src="https://storage.googleapis.com/prd-ai-asset/2022/02/a8b8d5d0-image4-1024x683.jpg" alt="" />
Yamada: And it also includes data from Merpay and Mercari Shops. Now that the data infrastructure has been established, we can say that the timing is right for a variety of interesting challenges. We need to think creatively about what we can do with the Mercari data and Mercari services.
At the same time, we’re an organization that takes a medium- to long-term view and can tolerate failure. For example, we can work over long spans of time to regenerate the data and build a data infrastructure. And in terms of human resources, we’ve hired many new graduates from around the world, including the Indian Institute of Technology (IIT). I think this kind of environment is a natural climate for nurturing ML engineers. What these things have in common is they don’t produce immediate results. It's about thinking in the medium to long term. That's why I want to hunker down and get to work on all of these issues, including AI ethics.
Mercari's use of AI began in 2017. In addition to improving the UI/UX, AI has contributed to the development of features that make Mercari safer and more secure. 
Related article: <a href="https://ai.mercari.com/en/articles/ai/suadd_kimuras_1/">AI Listings, Image Search… CEO Shintaro Yamada and Director Shunya Kimura Look Back at Mercari's AI Journey</a>
In the future, we will continue to seek new innovations with the aim of contributing to a recycling-oriented society through AI. Mercari's AI team will continue to take on new challenges.
Photo : Shun Nakayama (<a href="https://shunnakayama.myportfolio.com/">Web</a>/<a href="https://twitter.com/shunnakayama_">Twitter</a>) 
Text : Ayumi Iga (<a href="https://twitter.com/aymmm_i">Twitter</a>)

Contributing to a Circular Economy : CEO Shintaro Yamada and AI Director Shunya Kimura Discuss the Challenges that Mercari’s AI Technology Needs to Overcome

During 2017-2018, Mercari released a variety of features powered by AI technology, including AI listings, personalized timelines, and image search. These are still core features of Mercari. What was the process leading up to that? Shintaro Yamada, Representative Director &amp; CEO, and Shunya Kimura, Director of AI &amp; Search, look back at the history of Mercari's active use of AI, the challenges they faced along the way, and the organizational transformation that resulted.
<h2>Mercari has always had a data-driven decision-making culture</h2>
<img src="https://storage.googleapis.com/prd-ai-asset/2022/01/6dd58f3a-image1-1024x683.jpg" alt="" />
<div style="text-align: center;">Shunya Kimura, Director of AI & Search </div>
Shunya Kimura: At Mercari, we’ve been focusing on the use of AI for some time now. When did you decide to make AI the focus?
Shintaro Yamada: Mercari was originally a company with many employees who had worked in game development. I was also at a company called Zynga, and when I founded Mercari, we had many employees who joined us from DeNA and mixi in particular. From the days when we were creating game services for feature phones through the transition to smartphone culture, we all felt the importance of continuing to be data-driven. That's why we've been doing A/B testing and data-driven decision-making and product development.
Kimura: Was that inspired by your own first-hand experience?
Yamada: When I was at Zynga, we actively used A/B testing to select button colors and try out new game systems in the course of system development. By implementing the PDCA (Plan-Do-Check-Act) cycle, we were able to see the impact on our business, such as improved customer retention and increased sales. When we released Mercari in 2013, we didn't even have BigQuery yet, but we were doing data enhancement from the very beginning. Utilizing those systems, we saw ways to utilize more value-added data, including AI and machine learning.
Kimura: When did that happen?
Yamada: I remember it was around 2015-2016 that the world began to take notice of the active use of deep learning in AI. We asked ourselves if we might be able to use AI to develop a variety of features. We started development, and in 2017, we started developing AI listings and other products that incorporate machine learning.
<h2>The catalyst for the use of AI was witnessing the use of data by Chinese companies</h2>
Kimura: What inspired you to actively promote the use of AI? Was it your own sense the power of AI?
Yamada: Looking back at the developments within the company, the flow of events was as I just described, but personally, I was really struck by how the various services in China were using AI. At the time, Merpay was in its start-up phase, and we began getting many reports within the company from our employees who went to research FinTech services not only in China but also other countries. In the summer of 2017, I visited China with Aoyagi (Naoki Aoyagi, CEO of Merpay and Mercoin) before he joined the company.
<img src="https://storage.googleapis.com/prd-ai-asset/2022/01/1581af58-image5-1024x683.jpg" alt="" /> 

<div style="text-align: center;">Shintaro Yamada, Representative Director &amp; CEO of Mercari</div>

ByteDance, which would later become famous in Japan for TikTok, developed a personalized news service using deep learning called TouTiao. It's a service similar to SmartNews in Japan. Meituan developed Dianping, a gourmet food site similar to Japan's Tabelog site, and then expanded into food delivery and ride sharing. In each case, they progressively applied AI. Baidu's search engine was also amazing.
What they had in common was they collected data from users, placed that data at the center of their business strategy, and developed their services by thinking about how they could use it and what they could do with it. In the case of Meituan, they had just entered the ridesharing market using their data logged from food delivery bikes and other data.
At the time, seeing this company that was gaining momentum in China that was a data company making good use of data, it made me realize that it was possible to develop a range of businesses in this way. And looking at WeChatPay, Alipay, and so on, I could see that there was a new form of finance ahead.
Kimura: Indeed, the use of data by Chinese companies back in 2017 was quite a shock.
<h2>Start of implementation of core Mercari features “personalized timeline” and “AI listings”</h2>
Kimura: The first step in the use of AI at Mercari was the development of the &quot;AI listing&quot; feature. This is a feature that, when you take a picture of a product, suggests information about it and even estimates the price. It made an impact on the world, and some people still find this feature so useful that they say it’s one of the best things about Mercari. How did you feel about it at the time?
Yamada: The first thing that struck me was simply how far AI had come. It was a way to take a picture of a product with your phone, identify the products and brand, and even estimate the price. When I heard an overview of the feature, I honestly thought that we would not be able to achieve that level of accuracy. However, we took suggestions from the product side and proceeded with the implementation. When I tried out the actual feature, I was really surprised because the accuracy was higher than I expected. The image recognition technology is implemented correctly and incorporated into the feature in a way that is consistent with the Mercari experience. In fact, the completion rate from the time the listing screen is displayed to the time the listing is completed has also clearly improved. After this series of experiences, I felt the need to invest more in technology in the future.
<iframe loading="lazy" width="560" height="315" src="https://www.youtube.com/embed/NHxc2m3SEL8?controls=0" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>

<div style="text-align: center;">Video introducing the AI listing feature</div>

After that, we implemented a “personalized timeline” feature that personalizes the order of products displayed on your timeline. At the time, Mercari just displayed all the product categories in order of recency, but we released a feature that estimates categories of interest from the customer's browsing history and puts them on the timeline. This feature was very effective and had a significant impact in terms of total transaction value.
Kimura: The great thing about AI is that it can make suggestions that are one step ahead of, or at an angle to, what you might want to buy.
Yamada: I think what's nice about AI is that you don't need to manually construct a rule-based display order. The system is able to handle and process a wide range of data and automatically derive the most relevant output. How do you change the numbers in a particular product category for a particular customer segment? How do you adjust when the response is not good? The beauty of AI is that it can do that kind of complicated automation. I'm also inspired by the personalized services that TouTiao and Meituan have come up with.
<img src="https://storage.googleapis.com/prd-ai-asset/2022/01/2f7d3cdf-image4-1024x683.jpg" alt="" />
<h2>In a rapidly expanding organization, don’t try to march in lockstep, just create synergies where you need them</h2>
Yamada: When you first joined the company, what was the organizational situation like?
Kimura: In July 2017, when I joined the company, there weren’t as many engineers as there are now. We didn't even have an AI team yet. AI-driven development is now done by ML engineers, but back then it was done by back-end engineers. I remember that the AI team was formed in early 2018. We assigned AI engineers to various projects such as violation detection, AI listing, photo search, price suggestion, and for the US version of Mercari, weight estimation and shipping method recommendation.
At the time, we had created Merpay and were growing as an organization, and we kept the projects separate. Our ML engineers were split into different organizations such as Merpay and US version of Mercari. Under this arrangement, we didn’t try to align our AI development efforts, but proceeded with development in each individual organization. Rather than working too closely together, which slows down the development process, each organization works independently, loosely connected to the others. We emphasize speed while creating synergies where we need them. In this context, where Mercari's services are diversifying and globalizing, each organization is doing its own development using AI. I find that really interesting.
<img src="https://storage.googleapis.com/prd-ai-asset/2022/01/3caec9fc-image3-1024x683.jpg" alt="" />
<h2>The data management project that became a major task for advancing the use of AI</h2>
Yamada: The headcount has increased considerably since then. We now have several hundred engineers. There was a time when the A/B testing framework was full and we started to have conflicts. From that time on, I felt that automation of A/B tests was essential. I also felt it was extremely important to have the data acquisition system organized before reaching that phase.
At the time in 2019, Mercari had already grown into a service used by many customers, but we had challenges such as a lack of consistency in how data is obtained and where data is stored. After that, we started a project to revamp how we obtain data.
Kimura: We first tackled the development of a proper data infrastructure around 2019. As background, amid the acceleration of data-driven decision-making and rising expectations for AI development, the need to improve the accuracy and timeliness of data was growing across the company. In order to strengthen AI utilization and data-driven decision-making, it became necessary to reorganize the structure of the data Mercari was obtaining and the ways we were obtaining it.
<img src="https://storage.googleapis.com/prd-ai-asset/2022/01/222f69c7-image6-1024x683.jpg" alt="" />
At the time, the system was designed to make data storage quite free and easy, because we were focusing on flexibility in analysis. As a result, we ended up with large amounts of data stored in databases, making data maintenance and management increasingly difficult. There is also a debate going on about data management, with each stakeholder having their own opinions from a legal, engineering, and analytics perspective. Meanwhile, the development of Merpay started, and the situation became quite complicated.
This is still a problem that needs to be tackled company-wide, and there are a large number of stakeholders. Jeff, the incoming CEO of Mercari Japan, organized the situation, moved forward with the implementation, and gained company-wide consensus. Since then, we’ve unified how data is obtained and where data is stored, and we now have a more robust data platform that can store data in real time. Furthermore, a project to improve data governance is currently underway company-wide.
<img src="https://storage.googleapis.com/prd-ai-asset/2022/01/a7895e3d-image2-1024x683.jpg" alt="" />
Related article: 
<a href="https://ai.mercari.com/en/articles/ai/suadd_kimuras_2/">Contributing to a Circular Economy : CEO Shintaro Yamada and AI Director Shunya Kimura Discuss the Challenges that Mercari’s AI Technology Needs to Overcome</a>
In this article, the two discuss Mercari's vision of AI ethics and a concrete path forward to achieving a recycling-oriented society.
Photo : Shun Nakayama (<a href="https://shunnakayama.myportfolio.com/">Web</a>/<a href="https://twitter.com/shunnakayama_">Twitter</a>) 
Text : Ayumi Iga (<a href="https://twitter.com/aymmm_i">Twitter</a>)

AI Listings, Image Search… CEO Shintaro Yamada and Director Shunya Kimura Look Back at Mercari’s AI Journey

The &quot;MERCARI GEARS Lecture Series&quot; is a video series for engineers that introduces the direction that Mercari Inc. and other companies in the Mercari Group are heading and the technical challenges they will tackle in the future.
In this edition of MERCARI GEARS, Mercari's Customer Relationship Management (CRM) ML team* introduces how machine learning and causal inference can be used to provide a better marketing experience and improve campaign effectiveness.
*Now renamed the Marketing Data Science team
<center> 
<iframe loading="lazy" width="560" height="315" src="https://www.youtube.com/embed/g1r67WFVKUU" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe></iframe> 
Optimizing Marketing Campaigns using Artificial Intelligence [AI@Mercari #1]</center>
The efficiency of Mercari's coupon distribution was greatly improved by incorporating machine learning models into the segmentation definition of coupon distribution. Mercari's AI team is putting AI into practice to create a better customer experience.
Related projects:. 
<a href="https://ai.mercari.com/en/projects/eaf0a7f389/">Applying Uplift Modeling to Increase Customer Satisfaction and Streamline Marketing Efforts</a>

AI Application to CRM – Providing Better Marketing Experience and Improving Campaign Effectiveness

MERCARI GEARS Lecture Series&quot; is a video series for engineers that introduces the future direction and technical challenges that Mercari Inc. and other companies in the Mercari Group will tackle in the future.
In this edition of MERCARI GEARS, we focus on the moderation system developed by Mercari's TnS team (Trust and Safety team).
<center> 
<iframe loading="lazy" width="560" height="315" src="https://www.youtube.com/embed/STmIHmDCnfI" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe></iframe> 
Violation Detection Using Machine Learning at Mercari [AI@Mercari]. 
</center>
The moderation system is a system used by the Customer Success (CS) team to provide a trust and safety marketplace, and uses machine learning to detect items that may violate market policies.
To achieve a safer marketplace, we have devised a system architecture for the moderation system, separate models for each market policy, and request processing in coordination with the CS, which is explained in the video.
Related projects : 
<a href="https://ai.mercari.com/en/projects/auto-content-moderation-for-trust-safety/" title="Auto Content Moderation for Trust &amp; Safety">Auto Content Moderation for Trust &amp; Safety</a>

For a Trust and Safety Marketplace, Machine Learning for Violation Detection

The Mercari AI team’s research “Strategic Coupon Allocation for Increasing Providers’ Sales Experiences in Two-sided Marketplaces” has been accepted at the KDD2024 TSMO 2024 Workshop

mercari AI’s research “LLMs as an Interactive Database Interface for Designing Large Queries” accepted to HILDA Workshop in SIGMOD 2024

How have the organizational challenges of mercari  ML team been solved? A chat with the engineers who launched the company-wide Guild

Mercari AI team presented in NLP2024 “A related document retrieval system using past queries”

Mercari ML Engineers Chingis Oinar and  Teo Narboneta Zosa will give a talk at FOSSASIA Summit2024

Blogs by mercari AI team, delivering information about AI technology and organization

The Mercari AI team’s research “Strategic Coupon Allocation for Increasing Providers’ Sales Experiences in Two-sided Marketplaces” has been accepted at the KDD2024 TSMO 2024 Workshop

mercari AI’s research “LLMs as an Interactive Database Interface for Designing Large Queries” accepted to HILDA Workshop in SIGMOD 2024

How have the organizational challenges of mercari ML team been solved? A chat with the engineers who launched the company-wide Guild

Mercari AI team presented in NLP2024 “A related document retrieval system using past queries”

Mercari ML Engineers Chingis Oinar and Teo Narboneta Zosa will give a talk at FOSSASIA Summit2024

The Mercari AI team’s research “Strategic Coupon Allocation for Increasing Providers’ Sales Experiences in Two-sided Marketplaces” has been accepted at the KDD2024 TSMO 2024 Workshop

mercari AI’s research “LLMs as an Interactive Database Interface for Designing Large Queries” accepted to HILDA Workshop in SIGMOD 2024

How have the organizational challenges of mercari ML team been solved? A chat with the engineers who launched the company-wide Guild

Mercari AI team presented in NLP2024 “A related document retrieval system using past queries”

Mercari ML Engineers Chingis Oinar and Teo Narboneta Zosa will give a talk at FOSSASIA Summit2024

The Mercari AI team’s research “Strategic Coupon Allocation for Increasing Providers’ Sales Experiences in Two-sided Marketplaces” has been accepted at the KDD2024 TSMO 2024 Workshop

mercari AI’s research “LLMs as an Interactive Database Interface for Designing Large Queries” accepted to HILDA Workshop in SIGMOD 2024

How have the organizational challenges of mercari ML team been solved? A chat with the engineers who launched the company-wide Guild

Mercari AI team presented in NLP2024 “A related document retrieval system using past queries”

Mercari ML Engineers Chingis Oinar and Teo Narboneta Zosa will give a talk at FOSSASIA Summit2024

Articles

Mercari AI’s research paper—”Textual Content Moderation in C2C Marketplace”—accepted to ACL2022 ECNLP Workshop

Attempt to improve item recommendation accuracy using Item2vec

Edge AI – Realtime Item-Scan for iOS and Web

How we optimized our Machine Learning Training Infrastructure Costs @ Mercari US

Contributing to a Circular Economy : CEO Shintaro Yamada and AI Director Shunya Kimura Discuss the Challenges that Mercari’s AI Technology Needs to Overcome

AI Listings, Image Search… CEO Shintaro Yamada and Director Shunya Kimura Look Back at Mercari’s AI Journey

AI Application to CRM – Providing Better Marketing Experience and Improving Campaign Effectiveness

For a Trust and Safety Marketplace, Machine Learning for Violation Detection