Alibaba Cloud Summit | AI Search to Accelerate Scenario Innovation in Online Education Industry
Search is a powerful tool for online education enterprises to acquire traffic
As of December 2020, according to the statistics of the TOP10 monthly activities in the education industry, there are as many as 5 software with the ability to search questions. As a product capability, photo search questions can help customers acquire a large number of users and traffic, thereby providing liquidity for other products. Because of this positioning, the overall accuracy and search efficiency of photo search questions have become crucial points, so Open Search has made a lot of customized optimizations for this Alibaba Cloud Summit.
Features of Educational Search Questions
For the education search business scenario, three major characteristics are summarized:
The first point is the massive question bank, the education question bank belongs to the level of tens of millions or even 100 million, and continues to grow; at the same time, there is an obvious peak phenomenon in the question search business, such as seven or eight o'clock in the evening, the last day of holidays, when the question is searched There will be very high QPS peaks; search latency will seriously affect the user experience Alibaba Cloud Summit.
The second point is rich in scenes. The scenes covered by photo search questions are becoming more and more abundant, including different age groups. For example, the lower grade search questions mainly focus on taking pictures, reading pictures, literacy or connecting questions, and topics that require more picture information; also include Different disciplines currently support more than ten disciplines, so rich scenarios will bring greater challenges to the search effect Alibaba Cloud Summit.
The third point of algorithm requirements, the product form of photo search questions generally only shows the results of TOP3 or TOP5. It is precisely because of this setting that accuracy is very important for photo search questions. At the same time, photo search questions will also involve multi-mode state and multi-language processing capabilities to meet the needs of image and text search and multi-language processing Alibaba Cloud Summit.
Open Search Educational Questions Scheme Architecture
Alibaba Cloud's open search solution for photo search questions, when users take pictures and identify texts by OCR, after being processed by the open search engine, the results of TOP3-5 will be returned to users for display, and data security is strictly guaranteed for enterprise question bank data. and privacy.
Educational search algorithm ability
Query analysis algorithm optimizes the complete processing flow
Education industry word segmentation and subject category prediction
There are two major difficulties in word segmentation in the photo search scene Alibaba Cloud Summit . The first point is that spaces are missing after OCR recognition of English questions. As you can see from the first picture on the left, even for very long English texts without spaces, the model can be very accurate. segmentation. The second difficulty is the segmentation after the formula of the mathematical question. The second picture on the left shows that the mathematical symbols are all correctly segmented.
The category prediction corresponds to the prediction of the subject and the prediction of the question type in the photo search scene. We combine the pictures and the text information after OCR recognition to make multi-modal prediction, thereby improving the accuracy of the search Alibaba Cloud Summit.
Multiplex Recall Sorting Technology
Due to the particularity of the photo search business scenario, Open Search also introduced a multi-channel recall sorting technology.
Why do multiple recalls?
Compared with traditional web page or e-commerce search, there are obvious differences between educational photo search questions. The first point is that the search query is very long, and the second point is that the search query is the text obtained after the photo OCR recognition, and the key TERM recognition error If so, it will seriously affect the recall sorting Alibaba Cloud Summit.
There are two traditional plain text query solutions, the first is OR logical query, and the second is AND logical query. The AND logical query is based on the analysis of the Query module optimized and customized for the education field that we just mentioned, and the effect is greatly improved. Now The accuracy can be close to OR logic.
How to balance the computational cost of search and the accuracy of search Alibaba Cloud Summit?
The vector recall of text is introduced, and three points are optimized for the text vector recall technology,
The first point is that for the BERT model, we use the StructBERT developed by DAMO Academy, and customize it for the education industry. At the same time, the BERT model is compressed and accelerated Alibaba Cloud Summit.
The second point is that the vector retrieval engine adopts the Proxma engine developed by DAMO Academy, which surpasses the open source system in accuracy and speed.
The third point is that training data can be continuously accumulated based on customer search logs, and the effect will continue to improve.
As can be seen from the figure on the right, in the end, we can achieve very good results in the BERT model based on both sides. The accuracy exceeds the OR logic by 3%-5%, the overall number of recalled DOCs is reduced by 40 times, and the Latecy is reduced by more than 10 timesAlibaba Cloud Summit.
Knowledge Base Team
Knowledge Base Team
Knowledge Base Team
Knowledge Base Team
Explore More Special Offers
50,000 email package starts as low as USD 1.99, 120 short messages start at only USD 1.00