Core demands
1. First of all, Unilever, as one of the world's largest consumer goods companies, has cooperated with as many as 200000 offline suppliers, which is essential for the qualification review of new businesses. However, the traditional review process requires merchants to manually input the relevant information of business license in the system. Due to the large amount of information, there will be errors in the input of company name, registration number and other information. The manual review in the background may not only have errors but also have high costs. Therefore, there is an urgent need for an efficient and high accuracy solution to automate this tedious manual input link.
2. Unilever needs to interact with consumers during some special events to generate information about bills, so it needs to collect the pictures of bills uploaded by consumers. However, the pictures uploaded by consumers come from different channels of retailers, which leads to problems such as unclear pictures, out of center receipt subjects, and inconsistent commodity information formats. These problems will hinder the extraction of effective information after shopping.
3. In addition, there are many brands under Unilever, so how to more efficiently tap users' voice in social networks and quickly grasp the changes of users' emotional tendencies is also an urgent business problem.
Solution
1. Solution for qualification review of business license:
After using the business license recognition function in Baidu Brain character recognition, Unilever's internal AI laboratory (D-Lab) applies this technology to the business review platform. Users only need to scan the code on WeChat to enter the specific H5 interface, and upload the business license image by taking photos to complete the review.
The optimized business qualification review process is as follows:
Step 1: Cooperative suppliers can access the specific H5 interface through WeChat scanning;
Step 2: The user uploads the picture of the business license by taking photos;
Step 3: Unilever calls Baidu Brain's business license identification interface to structurally extract the company name, legal person, address, validity period, certificate number, social credit code and other information;
Step 4: The audit platform will compare the identified results with the national effective enterprise industrial and commercial information database, and if the information is consistent, it will pass. If there is inconsistent approval or unrecognized information in the identification process, the merchant will be prompted to re verify and upload.
2. Solution for shopping ticket information extraction:
For the problem of extracting shopping receipt information uploaded by consumers during special activities, Unilever AI Lab (D-Lab) has built a shopping receipt information extraction system by combining Baidu Brain's general receipt recognition technology with D-Lab's self built knowledge map in e-commerce field.
The specific implementation process is as follows:
Step 1: Unilever AI Lab (D-Lab) classifies the junk images in a round to eliminate non junk images;
Step 2: use Baidu Brain's ability to identify general notes to extract relevant text information;
Step 3: Through the D-Lab's self built knowledge map of e-commerce field, the second extraction of commodity information and geographical location information is carried out for the text identified by the general bill, so that it can accurately locate the product brand and category, store geographical location, store region and other information appearing in the ticket image.
3. Solution for public opinion monitoring:
For the problem of real-time understanding of product word-of-mouth and intelligent monitoring of public opinion, Unilever uses Baidu Brain's comment extraction technology and sentiment analysis technology to solve:
With the help of Baidu Brain's comment opinion extraction technology in the food field, the food language corpus is mined and extracted to help analysts find the opinion information in sentences faster, shorten the time for manual participation in parsing, and improve efficiency.
While sentiment analysis is based on deep learning technology and Baidu big data. For Chinese text with subjective description, it automatically judges the emotional polarity category of the text and gives corresponding confidence. By analyzing the emotional orientation of the comment corpus in social media, the efficiency of brand market analysts on the business demand side in searching for high-quality and low-quality content has been improved, and the change of users' emotional orientation has been quickly grasped.
With the help of Baidu's above technologies, we can better understand the emotional division of consumers when discussing different topics, help brands better understand the characteristics and reputation of products, and help optimize brand marketing strategies and research and development of new products.