Market Size in the World of Data Labeling

1. What is Data Labeling?

Data labeling, which is essential for "supervised learning," refers to the process of assigning labels to data, as the term suggests. But what does it mean to label data? Here, we will explain the mechanism of data labeling.

1-1. Definition and Characteristics of Data Labeling

Data labeling is the process of marking the objects within data that you want the AI to recognize. There are various ways to apply these labels. For example, in image labeling, there are methods such as "bounding boxes," which enclose the object in a rectangle; "keypoints," which mark specific positions on the object; and "segmentation," which involves coloring in the contours of the object. For text labeling, methods include selecting the target parts within a sentence (such as words, sentences, or paragraphs) by underlining or highlighting, or labeling the entire text as a whole. Although the purposes differ, the comment function in Word documents can also be considered a type of labeling. The assigned labels are given types or information (called classes) that indicate what the object is, such as "car," "person," or "traffic light." The data labeled in this way becomes training data and is used for AI learning.

1-2. What is the difference from annotation?

Many of you may have seen the explanation that involves performing "annotation" to create training data. Our company also has a blog that explains "annotation."

>>What is annotation? Explanation from its meaning to its relationship with AI and machine learning.

 

Although "data labeling" and "annotation" may seem to have different meanings at first glance, they can be considered the same in terms of creating training data. However, as English words, they also have distinct meanings. Since methods of AI learning, such as deep learning, have primarily developed in English-speaking regions (especially the United States), let's take a look at how these two terms are treated in English.

1-3. Differences in the Use of Terms Between Japan and Overseas

In English, "annotation" originally means "to add notes or comments to a text." For example, it refers to marking relevant parts of a text with symbols like "*" or underlining and adding notes. When creating training data, similar work is done by marking data and assigning classes, which is why it came to be called "annotation." On the other hand, "data labeling" also refers to the same kind of work as "labeling," such as attaching price tags or labels to products to provide information like product names and prices, and is therefore also used when creating training data.

 

When you search the word "annotation" alone in English on the web, content related to writing and proofreading, such as "how to add annotations to text," tends to rank higher. In contrast, AI-related content ranks higher for "data labeling," so it seems that "data labeling" is more often used to refer to the creation of training data. For example, in the United States, some training data creation tools have names like "labelimg" and "labelme," which evoke labeling. Also, in the blog of a company called Superannotate that provides online tools, the expression "be labeled" frequently appears.

 

That said, both "data labeling" and "annotation" can be considered the same term referring to the task of creating training data. However, "data labeling" tends to evoke a more concrete action, so it is used more frequently at the level of tool operation instructions and training data work manuals. Also, the workers who create training data are commonly called "data labelers."

 

On the other hand, in Japan, the process of creating training data is often referred to as "annotation." The workers who create the training data are commonly called "annotators" in Japan.

2. Market Size of Data Labeling

Various reports have been released regarding the market size of data labeling, and the figures for market size differ among them. However, all reports indicate that the data labeling market is on an upward trend. Here, we will refer to a report released by Markets and Markets in February 2023 and an analyst note released by UBS in July 2023 to explain the market size of data labeling.

2-1. Current Market Growth Rate

As of 2022, the global market size for data labeling is 800 million USD. According to the report, it is expected to expand to 3.6 billion USD by 2027, which represents an average annual growth rate of 33.3%.

2-2. Future Trends and Forecasts of the Data Labeling Market

The market size for data labeling continues to show an expanding trend, which has been reported in other reports besides Markets and Markets. Factors contributing to this expansion include the broadening scope of AI applications due to advancements in AI technology. Among these, a significant factor for growth is the increasing demand in medical imaging. It is anticipated that there will be applications in image diagnostics that do not require medical professionals, the introduction of medical robots, and document searches for various medical records, papers, and documents issued during new drug development (where natural language processing technology enhanced with medical terminology, in addition to AI OCR, will be necessary).

2-3. Relevance to Next-Generation Technologies and Industry Development

The financial services division of the major Swiss bank UBS raised its long-term AI demand forecast in an analyst note released on July 25, 2023, from an average annual growth rate of 20% over five years starting in 2020 to a new average annual growth rate of 61% over five years starting in 2022. This forecast is considered to take into account the rapid spread of generative AI usage, represented by ChatGPT. Such developments related to AI are recognized not as a transient growth like an AI bubble, but as a long-term trend.

 

>>UBS Forecasts AI Demand to Grow at an Average Annual Rate of 61% from 2022 to 2027

 

Data labeling is expected to play an important role in the learning of AI, even in the field of generative AI, just as it has in the past. Considering the expansion of its use in industries such as healthcare mentioned earlier, it can be said that the scope of AI requiring data labeling will continue to expand in the future.

>>What AI and Machine Learning Can Do: 12 Use Cases by Industry.

2-4. The Impact of Increased Demand for Data Labeling

The increasing demand for data labeling will create new job opportunities. Data labeling can be done remotely depending on the security level, allowing for a wide range of talent to be secured without being limited by region or time zone.

 

On the other hand, if the data contains information related to privacy or security, there are concerns about leakage or exposure if the data is not handled in an appropriate environment. This is especially relevant for medical-related data labeling, where demand is expected to increase, and such data is often handled. In particular, data labeling vendor companies are increasingly required not only to focus on building secure remote work environments to ensure thorough security management but also to provide security rooms for on-site handling and to be able to accommodate on-site presence at client locations.

2-5. Use Cases of Data Labeling

Case 1: Kajima Corporation

The company was advancing the AI development and implementation of an on-site management system using webcams and other tools to improve productivity and working conditions. However, acquiring the labeled data needed to train AI models, which requires enormous time and cost, became a technical and time-related bottleneck. Since they could not allocate enough effort to model design and development, which should have been their main focus, outsourcing data labeling to an external vendor allowed them to secure time for model design and accelerate AI development more efficiently.

Case Study 2: ORIX Corporation

In developing AI tools to streamline the processing of document data such as invoices, the company outsourced to vendors, enabling efficient one-stop handling from data collection to data labeling even for hard-to-collect items like overseas invoices and receipts. As a result, the development and accuracy improvement of AI tools proceeded smoothly.

Case 3: Harvard Medical School

At this research institute, they are developing mouse behavior analysis models in the study of neural mechanisms. Until now, researchers themselves manually labeled mouse videos, which consumed an enormous amount of time. However, by leveraging the support of an outsourced vendor, they succeeded in reducing labeling time by several weeks. Being able to obtain a large volume of labeled data in a short period has made it easier to create high-performance machine learning models and allowed researchers to allocate more time to areas requiring specialized knowledge that contributes to advancing their research.

Our Company's Data Labeling Use Cases

SCSK Corporation — Achieving Efficiency, Quality Improvement, and Cost Reduction through Annotation Outsourcing ~Improving AI Development Work Efficiency through Division of Labor, Contributing to Shortening Development Periods~

Sumitomo Heavy Industries, Ltd. — Ensuring the accuracy and reliability of machine learning systems by outsourcing fast and accurate annotation work

3. What are the benefits of outsourcing data labeling services?

Data labeling currently requires human intervention. Moreover, since labels must be applied to a vast quantity of data ranging from thousands to hundreds of thousands, it often takes weeks to months to complete. In companies engaged in AI development, there are cases where development engineers handle this labeling, but this can encroach on the time needed for their primary development tasks. Here, we will explain the benefits of outsourcing data labeling tasks.

3-1. Possibilities for Cost Reduction and Efficiency Improvement

Data labeling, unlike AI development tasks, does not require programming skills or specialized knowledge in AI engineering. Furthermore, it is a task that can take up a significant amount of time in the AI development process. If engineers perform this task, it incurs costs unrelated to the core development work. Additionally, even if a company secures personnel for labeling, it can become a waste if labeling tasks do not arise. Therefore, considering the effort involved in securing personnel and managing data labeling, outsourcing to specialized vendors instead of performing labeling in-house can be said to be the greatest advantage in terms of cost reduction and operational efficiency.

3-2. Utilization of Specialized Knowledge and Skills

Data labeling does not require the expertise of engineering and other skills necessary for AI development, but to maintain data quality and meet deadlines with high productivity, appropriate management skills, as well as expertise and know-how related to labeling, are essential. By outsourcing to external vendors, you can leverage the unique expertise and know-how specific to data labeling.

3-3. Points to Consider When Outsourcing Data Labeling Services

When outsourcing, it is recommended to keep the above benefits in mind and talk with multiple vendors, including those who have a track record in data labeling that matches your company's AI development goals, ensure quality, and implement security measures. For your reference, please also see the following blog from our company.

>>How to Outsource Annotation Work? 7 Tips

3-4. Precautions to Avoid Failure When Outsourcing Data Labeling

An important aspect when proceeding with outsourcing is management after placing the order. Even after exchanging estimates and specifications and placing the order, unexpected cases or discrepancies in judgment criteria may arise once the actual work begins, potentially affecting data quality. To promptly correct quality degradation caused by such misunderstandings, regular meetings with the vendor and a system to reliably share changes are indispensable. Additionally, it is crucial to have a structure that can flexibly adjust personnel and schedules when the project expands. It is important not to choose solely based on price or cost, but to comprehensively evaluate and consider vendors including their responsiveness, flexibility, and information-sharing systems.

4. Achievements in Human Science

We would like to introduce interview articles featuring feedback from our clients based on our past achievements. Please take a look.

>>Achieving Fast and Accurate Annotation Work through Outsourcing: Ensuring the Accuracy and Reliability of Machine Learning Systems (Sumitomo Heavy Industries, Ltd.)

5. Human Science Data Labeling, LLM RAG Data Structuring Outsourcing Service

Over 48 million pieces of training data created

At Human Science, we are involved in AI model development projects across various industries, starting with natural language processing, including medical support, automotive, IT, manufacturing, and construction. Through direct transactions with many companies, including GAFAM, we have provided over 48 million high-quality training data. We handle a wide range of training data creation, data labeling, and data structuring, from small-scale projects to long-term large projects with a team of 150 annotators, regardless of the industry.

Resource management without crowdsourcing

At Human Science, we do not use crowdsourcing. Instead, projects are handled by personnel who are contracted with us directly. Based on a solid understanding of each member's practical experience and their evaluations from previous projects, we form teams that can deliver maximum performance.

Generative AI LLM Dataset Creation and Structuring, Also Supporting "Manual Creation and Maintenance Optimized for AI"

Since our founding, our main business and service has been manual creation, and now we also support "the creation of documents optimized for AI recognition" to facilitate the introduction of generative AI for corporate knowledge utilization. In sharing and utilizing corporate knowledge and documents using generative AI, it is currently difficult to achieve 100% accuracy with tools alone. For customers who absolutely want to leverage their past document assets, we also support the structuring of document data. Leveraging our unique expertise in various types of documents, we provide the optimal solution.

Secure room available on-site

Within our Shinjuku office at Human Science, we have secure rooms that meet ISMS standards. Therefore, we can guarantee security, even for projects that include highly confidential data. We consider the preservation of confidentiality to be extremely important for all projects. When working remotely as well, our information security management system has received high praise from clients, because not only do we implement hardware measures, we continuously provide security training to our personnel.

In-house Support

We also provide personnel dispatch services for annotation-experienced staff and project managers who match our clients' tasks and situations. It is also possible to organize teams stationed at the client's site. Additionally, we offer broad support for clients' challenges related to annotation and data labeling, including training for client workers and project managers, selecting tools according to the client's situation, designing automation and work methods, and building optimal processes to improve quality and productivity.