In our daily life, there are more or less accounting situations, so as to analyze our income and expenditure and consumption habits, to help ourselves reduce unnecessary expenses, optimize financial decisions, allocate funds reasonably, and reduce financial pressure and unnecessary waste.
But the act of bookkeeping itself is a bit more troublesome. Although there are many APP applications to help us keep accounts at this stage, and we no longer need to write records like before, there are still many problems. For example: every record has to be manually selected and classified, which is a waste of time in the long run. In addition, if you want to keep accounts automatically, you have to authorize the permission interface of the payment application to the accounting program, which will cause security risks to property.
Coincidentally, in the recent "China College Student Service Outsourcing Innovation and Entrepreneurship Competition", some college student teams provided their solutions to the bookkeeping problem.
First of all, let me introduce what is the China University Student Service Outsourcing Innovation and Entrepreneurship Competition
The China University Student Service Outsourcing Innovation and Entrepreneurship Competition is a national competition derived in response to the country's relevant strategic measures and calls for encouraging the development of the service outsourcing industry and strengthening the training of service outsourcing talents. National competition in the industry field. Especially in this competition, a total of 8006 teams from 803 national colleges and universities signed up for the competition, and the number of registered teams hit a new high! It can be said to be a very influential event.
The content design of this competition fully focuses on the practical problems of technology and management in the development of enterprises, and is more closely integrated with the industry. Among them, intelligent character recognition technology is one of the technologies that the competition focuses on.
As a leading technology company in the field of intelligent text recognition at home and abroad, based on its own cognition in the industry, Hehe Information issued a "product collection order" to the majority of students from the topics of "innovative application of personal financial management based on intelligent text scenarios". ", relevant competition questions attracted nearly 300 teams from more than 70 colleges and universities across the country to actively participate, and many excellent works emerged.
In particular, the "Almighty King of Bookkeeping" developed by Central South University impressed me deeply. This application is very effective in solving the pain points of current industry status bookkeeping:
Their team first conducted research on a large number of users, and analyzed that automatic bookkeeping and picture recognition bookkeeping are more popular, especially among middle-aged users aged 45-60:
We also investigated the common bookkeeping apps on the market, and pointed out several major problems existing in them at this stage, especially the one that automatic bookkeeping leads to excessive collection of privacy, and made a full explanation:
In response to this problem, they also provide a way to enter billing information by identifying the bill picture in the APP for accounting :
This function seems to be very simple, but the actual difficulty is very great!
Although there are only two core steps in this function: bill image recognition and information extraction.
But it takes a lot of effort to do these two steps well. First of all, there are many types of bills. Second, if the bills themselves are not well preserved, there will be many wrinkles or unclear handwriting. Third, the complex shooting environment will lead to poor quality problems such as reflections and differences in light and shade. This will easily cause inaccurate or failed recognition during recognition.
In order to solve the problem of bill recognition, they used the intelligent text recognition service platform interface provided by Hehe Information to identify and preprocess bills. This interface supports recognition of many types of bills, whether it is invoices, train tickets, financial bills, etc. Both can provide high-precision recognition results:
Once the image recognizes the text, it needs to be preprocessed. This step usually includes removing punctuation marks, numbers and special characters, converting the text to lowercase, and word segmentation. They used jieba, a word segmentation tool library specially designed for Chinese text, to perform word segmentation, and then converted the words in the text into numerical vectors so that the computer can understand and process them. Finally, they performed text classification and information relationship on the billing information in the picture. Extract, extract the specific amount, location, store and other information, that is, Named Entity Recognition (NER)
What is Named Entity Recognition (NER)?
It refers to the identification of entities with specific meaning in the text, mainly including names of people, places, institutions, proper nouns, etc., and mark the words we need to recognize in the text sequence.
It is easy to understand with an example. For example, there is a piece of text:
Zhang San and I went to see Spider-Man yesterday, and it felt good. We want to see Avatar next week. Do you want to come with us?
We want to identify the information of the movie name in the above text, then we need to identify the content: Spiderman, Avatar.
For the bill recognition mentioned above, we need to extract the bill-related parts of the text information contained in the picture and exclude irrelevant information. This is a typical named entity recognition.
In order to deal with this problem, Central South University used Bert-Chinese derived from Google BERT (Bidirectional Encoder Representations from Transformers) as a pre-training model:
Various ticket type data are then fed into the model and trained. After the training is completed, fine-tune it, apply it to downstream tasks (such as bill category determination), and finally extract the bill information. The flow chart of the entire algorithm step is shown in the figure below:
The technical route used by their team is shown in the figure below:
I also used the "Bookkeeping Almighty King" app to test it, and the effect is also very good:
recognition result
In my opinion, the overall performance of the Central South University team is very impressive. It not only has insight into the phenomenon that "middle-aged and elderly people need to go through cumbersome operations to use bookkeeping applications", but also found the trend of "picture recognition bookkeeping is more popular" and Product optimization is carried out in a targeted manner, and the API of Hehe Information's intelligent image recognition module and receipt recognition is flexibly implemented in the accounting scene, and combined with the large model, the complex receipt information is converted into concise and efficient data in seconds Input, which is very valuable.
In addition, I also think that this type of application has good prospects and commercial value. The State Council issued "The 14th Five-Year Plan for National Economic and Social Development of the People's Republic of China and the 2035 Long-term Objective Outline", "New Generation Artificial Intelligence Development Plan", etc. The document also mentions that the in-depth application of artificial intelligence in the field of personal financial management is conducive to promoting the digitization of personal financial management, helping consumers to achieve reasonable arrangements for consumption, reliable protection of financial risks, and optimal management of money at a lower cost. Intertemporal configuration. With the strong support from the state, the track must have a bright future!
In the test conducted by the China Academy of Information and Communications Technology, the intelligent text recognition products of Hehe Information successfully passed all 7 basic functional index tests and 9 enhanced functional index tests, and received an "enhanced" rating. Its intelligent text recognition products performed well. performance and service maturity.
Taking the difficult performance test of certificates and bills as an example, in the face of complex scenes such as rotation, shadows, reflections, wrinkles, deformation, blur, multi-language, low pixels, and uneven lighting, Hehe Information's intelligent text recognition products are It has a high recognition accuracy rate, the character accuracy rates are 99.21% and 99.59%, and the field accuracy rates are 97.87% and 98.42%.
In fact, the function of Central South University to use Hehe Information's bill identification interface to identify bills is only a small part of the many functions of Hehe Information. In addition, Hehe Information also has many powerful functions and products, especially the scanning Almighty King, Business Card Almighty King and other intelligent text recognition products have served hundreds of millions of users in hundreds of countries and regions around the world.
Last year, I also used Hehe Technology's PS detection and moiré removal services, and the results were very good, especially PS detection, which has always been a difficult problem that many industries need to solve urgently, especially in insurance, finance, banking, etc. If the false and falsified information is approved, it may bring huge impact or even economic loss:
This year, we also saw that the Hehe Information team continued to optimize and upgrade the "black technology" of image tampering detection, and the application area was also expanded to "screenshot tampering detection": in addition to the originally supported document, certificate, certificate and other natural scene image recognition and detection, it also It supports the identification and detection of various screenshots such as transfer records, transaction records, chat records, etc., whether it is the "copy and move" image tampering method of "cutting out" key elements from the original image and then moving "paste" to another place, or "wiping "Delete", "reprint" and other methods, image tampering detection technology can "wisdom eye" to identify fakes!
It is not difficult to see that the products of Hehe Information are not only of high quality and full of diversity, but also can be applied in a wide range of fields.
Through this competition, we can also see that the works of modern students are no longer limited to application development under the traditional Internet thinking, but gradually develop into products that combine artificial intelligence and large models to innovate and create a new era. "Solving old problems with new technologies".
What can also be felt is that at this stage, the demand for talents of enterprises has changed from singleness to diversity. Talents with single knowledge can no longer meet the needs of the times, so cross-learning is becoming more and more important.
Another important purpose of holding the competition is to promote in-depth cooperation between the school and the enterprise in scientific research projects and personnel training, and to promote the collaborative innovation and development of industry, university, research and application. Therefore, the competition's scoring criteria for the entries are also very "simulation", involving technical resources and economic cost control, judgment on the creative prospects of the project, analysis of market demand, etc., covering commercial value, social application value and other aspects of evaluation .
The explosion of CharGPT and other generative AIs makes us clearly feel that the future must be the era of artificial intelligence, and the industry will also be eager for every talent who has a deep and unique understanding of professional academic fields and has the potential to build solutions.
At the closing ceremony of the competition, Du Jie, head of the human resources administration department of Hehe Information, introduced their company's talent training plan:
"The company expects to work with the new generation of young talents to jointly explore new technological scenarios. At every stage of the progress of the times, we need different new forces to create new possibilities." Du Jie said that at this stage, Hehe Information has passed " A series of talent cultivation programs and supporting sharing platforms such as the "Spark Plan" help scientific and technological youth strengthen their professional capabilities in practice. Find the fertile soil conveniently, so as to "grow flowers on the ground".
I believe their actions and the continuation of the competition will continue to influence more practitioners!