Shulou(Shulou.com)06/02 Report--

"We found eight professional lawyers and compared them with our AI robots. Our recall rate was comparable to that of eight lawyers, but our accuracy rate was higher."

At the recently concluded Hangzhou Yunqi Conference, Li Bo, a senior algorithm expert from Alibaba's Information Platform Division, was showing the audience Alibaba's automatic document review system. At present, this system has reached more than 98% accuracy in automatic identification and has been practiced in Ali's internal legal scene.

In fact, in addition to automatic document review system, artificial intelligence has been widely used in Alibaba intelligent HR, intelligent legal affairs and other fields.

How does Alibaba enable organizations to successfully complete digital transformation through AI and other technical means? At the enterprise application forum of Yunqi Conference on the 22nd, Li Bo unlocked the mystery of Ali's enterprise wisdom brain for everyone.

cdn.com/345ee2267f8725035d66b2207aab17ec07ac55af.jpeg">

As for "enterprise brain," a relatively complete expression was put forward by Inspur Group at the NPC and CPPCC in March 2018, that is, it is an intelligent open innovation platform for enterprises based on the integration of artificial intelligence, big data and other new IT technologies, assisting intelligent decision and business automation, driving intelligent business system and realizing personalized, customized and refined enterprise production and service.

"Business cloud, data integration, application innovation" are three steps to build an enterprise intelligent brain. Combined with the actual experience of Alibaba Information Platform, Li Bo believes that we can start from the following three best practices:

1. Break down information islands.

The "information island" here is mainly aimed at the internal data of the enterprise. One of the reasons is the imbalance of business development and the sequential relationship between the development of technical systems, which causes the two systems to use different data models when describing the same business concept, resulting in data failure. Second, a lot of data still exists offline, such as paper documents, legal documents, reimbursement bills and user behavior (such as whether the conference room is being used). At this time, we need to use AI technologies such as NLP and CV, combined with relatively low-cost Iot equipment, to break the information island and help us improve efficiency.

2. Deep integration into applications.

Integrate AI into applications in an industry +AI way. The internal operation of traditional industry organizations has been relatively mature, but there are a lot of manual work, low efficiency, and easy to make mistakes. AI intervention can better improve operational efficiency.

3. C2B migration.

That is, successful AI experiences in Class C are migrated to Class B applications. Li Bo believes that although there are differences between Class C and Class B applications, some good Class C experiences and technical practices can help effectively shorten the construction path of enterprise intelligent brain in Class B scenarios. This can be the key exploration direction of building enterprise intelligent brain in the next 5-10 years.

Combined with these three best practices, Alibaba has made some successful attempts in the fields of smart HR and smart legal affairs:

Intelligent Promotion Assistance-AI reduces artificial subjective bias

In HR promotion scenarios, if a team of a certain size wants to consider personnel promotion, it usually encounters the following two questions: 1. Which candidates have promotion potential. 2. Different candidates, who better meets the promotion criteria.

In the past, it was mainly the supervisor and HR who gave the answer, but there would be inevitable artificial bias in the middle. If AI is used to assist decision-making, it is possible to minimize subjective bias caused by human intervention.

Based on Alibaba Group's internal data, Alibaba has built an objective indicator system from four dimensions: performance and potential, precipitation and sharing, quality and output, input and efficiency, and built a machine learning model in this system to make auxiliary predictions for promotion.

For example, a candidate's promotion probability will be compared with the promotion criteria and previous promotion cases to help the supervisor or HR make decisions.

At present, AI decision-making of intelligent promotion assistance system mainly plays a role in nomination and evaluation stage. For entry-level and middle-level positions, the intelligent promotion assistance system has achieved a prediction accuracy rate of 98%. And it can cover 40% of the potential promotion population. For a large group like Alibaba, this number has played a considerable role in improving the efficiency of the enterprise.

Interviewer evaluation model--AI improves efficiency

In addition, Alibaba has created an interviewer evaluation model for HR promotion.

The interviewer's interview skills and maturity directly determine the efficiency and effectiveness of recruitment. However, unlike the promotion assistant model, the interviewer model lacks objective historical data.

In this regard, when constructing the interviewer evaluation model, it is necessary to choose the active learning method and combine manual modeling with machine modeling.

In the artificial modeling stage, expert experience is introduced deeply, and directional rule indexes are generated manually. Then, by manually labeling the results of the samples, the data is adjusted by backward extrapolation until the final interviewer evaluation is generated.

After having the manually labeled data, enter the machine modeling stage. In machine modeling, not only automatic models can be obtained, but also features can be mined from the data, such as: which characteristics of the interviewer match, what tendencies he has. These data features, in turn, assist in manual modeling and labeling.

With Active Learning, this interviewer evaluation model maintains an accuracy of over 90% for interviewers and covers 20% of interviewers. Although the 20% coverage figure may not be large in itself, it is enough to support the recruitment team to make corresponding operational adjustments such as interview skills training and follow-up for interviewers.

Intelligent flower name--AI catalyzes organizational culture with temperature

Flower name is Alibaba's unique culture, but also Ali this temperature of the organizational culture embodiment. However, due to the uniqueness of each employee's name, even if the employee's name will be retained, so new students find it difficult to get a name after entering the job.

Therefore, in the scene where the newcomer takes the flower name, AI provides such a function-intelligent flower name. It can randomly recommend names, specify keywords to retrieve names, and even name names based on description and interpretation preferences. For example, if you want the name to include the meaning of "leading the way," the intelligent name system will recommend names such as "Pioneer,""Pioneer," and "Pioneer." Since the launch of the intelligent roster system, the employee adoption rate has reached more than 60%.

The three AI systems mentioned above are mainly used in the HR field. In the legal field, Alibaba also handles some of its daily work through AI.

Automatic paperwork review-AI automates repetitive and cumbersome business processes

Automatic clerical agreement review is the AI robot mentioned at the beginning with 8 professional lawyers PK. It can automatically review potential risks in the protocol and give recommendations. This can further reduce platform risk. At present, the recognition accuracy rate of automatic document protocol review is about 98%, and 85% of violations can be detected.

In addition to agreement review, AI can also assist in contract form review. Including the consistency of contract text content, contract amount correctness check (such as whether the capital amount and lowercase amount are consistent), clause completeness check and serial number, typo check, etc. These are all in the daily work can greatly help the legal staff, from the daily lot of tedious work to free up, focus on more creative work.

Intelligent document entry_AI makes online and offline information seamlessly connected

A large number of legal documents, whether contracts, pleadings, or evidence, exist mainly in paper form. How to quickly input paper documents into the system is the most critical link to improve the overall work efficiency.

Smart document entry provides a solution to this legal scenario. It can not only automatically convert offline text to online, but also automatically extract key information.

After the paper document is scanned, the text content is identified through OCR, and at the same time, the text is analyzed and information extracted through mature NLP technology to extract some key fields, such as Party A and Party B. At the same time, the system can also classify the clauses, such as what types of clauses belong to and which clauses need to be focused on. At present, the accuracy rate of document extraction reaches 98%, and the accuracy rate of clause classification is about 94%.

These intelligently entered document information are very useful for subsequent search applications and BI statistical applications.

Smart Contract Search-AI makes contract search faster, more accurate and safer

Legal students have a large amount of contract retrieval in their daily work. In this scenario, the intelligent contract search function can achieve millisecond retrieval performance and retrieval response to ensure timeliness.

In addition, contract retrieval has high requirements for security and confidentiality. In the development and deployment stage, this system realizes a set of secret retrieval functions, which can effectively guarantee the security of data.

At present, according to the characteristics of legal documents, the system also realizes the customized retrieval and sorting process, so that the overall retrieval correlation is above 90%.

Enterprise intelligent brain is still a relatively emerging field. In the future, Alibaba will continue to deepen AI application and realize the digital transformation of enterprises.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.