AI Challenger 2018 text mining competition related solutions and code summary

AI Challenger 2018 It's almost over , Each track top The contestant has finished the code verification , Preparing for 12 month 18、19 Japan AI Challenger On the way to the final defense materials . This year AI Challenger When the dust is about to settle , Here is a collection of solutions and codes for text mining related tracks that are currently visible on the Internet , Welcome to add , At the same time thank you github, Thank you open source students .

Fine grained user comment sentiment analysis

Fine grained sentiment analysis of online reviews is very important for a deep understanding of businesses and users 、 Mining user emotion and other aspects has a vital value , And it is widely used in the Internet industry , Mainly used for personalized recommendation 、 Intelligent search 、 Product feedback 、 Business security, etc . In this competition, we provide a high quality mass data set , contain 6 Categories: 20 A fine-grained element of emotional orientation . Participants need to establish an algorithm according to the emotional tendency of the fine-grained elements , Sentiment mining for user reviews , The organizing committee will determine the prediction accuracy by calculating the error between the predicted value submitted by the contestant and the real value of the scene , Evaluate the submitted prediction algorithm .

It seems to be the hottest track ,Testa The submission team has 468 the , For details, please refer to the home page of the track ：https://challenger.ai/competition/fsauor2018

Relevant code or solution ：

1. official baseline, be based on SVM: sentiment_analysis2018_baseline
https://github.com/AIChallenger/AI_Challenger_2018/tree/master/Baselines/sentiment_analysis2018_baseline

2. be based on fastText Of baseline: AI Challenger 2018 Sentiment Analysis Baseline with fastText
2.1 https://github.com/panyang/fastText-for-AI-Challenger-Sentiment-Analysis
2.2 article ：AI Challenger 2018 Fine grained user comment sentiment analysis fastText Baseline

3. be based on SVM Fine grained emotion analysis : https://github.com/scruel/sentiment_analysis

4. The first 16 Name solution : https://github.com/xueyouluo/fsauor2018

5. The first 17 Name solution ：https://github.com/BigHeartC/Al_challenger_2018_sentiment_analysis

6. be based on Bert Attempts to ：https://github.com/brightmart/sentiment_analysis_fine_grain

7. ai challenger Competitions 1: Fine-grained Sentiment Analysis of User Reviews：
https://github.com/ShawnXiha/Fine-grained-Sentiment-Analysis-of-User-Reviews

8. Fine grained user comment sentiment analysis (0.70201)：https://github.com/pengshuang/AI-Comp
8.1 Related articles 1：AI-Challenger Baseline Fine grained user comment sentiment analysis (0.70201) Previous articles - You know
8.2 Related articles 2：AI-Challenger Baseline Fine grained user comment sentiment analysis (0.70201) Later chapters - You know

9. AI Challenger Fine grained user comment emotion analysis online 0.62: AI Challenger Fine grained user comment sentiment analysis - You know

Reading comprehension of viewpoint questions

Machine reading comprehension involves information retrieval 、 Text matching 、 Language understanding 、 Semantic reasoning and other technologies at different levels , Dealing with complex problems even requires the combination of world knowledge and common sense knowledge , Very challenging . In order to further promote the technological development in the field of machine reading comprehension , Provide researchers with benchmarks for academic exchanges and model evaluation , This competition will focus on the more complex in reading comprehension , We need to use the information of multiple sentences in the whole article to synthesize the viewpoint questions to get the correct answers . This competition will use the accuracy rate to score , As the main evaluation index . The organizing committee will adopt objective indicators , Combined with the performance of the defense , Algorithm model for comprehensively evaluating contestants .

For more information, please refer to the official homepage ：https://challenger.ai/competition/oqmrc2018

Relevant code or solution ：

1. official baseline: be based on pytorch Realize the thesis 《Multiway Attention Networks for Modeling Sentence Pairs》
opinion_questions_machine_reading_comprehension2018_baseline
https://github.com/AIChallenger/AI_Challenger_2018/tree/master/Baselines/opinion_questions_machine_reading_comprehension2018_baseline

2. take baseline Moved to python 3.6.6, Fixed a bug , Change the parameters to the highest accuracy by 0.70370：
https://github.com/dreamnotover/oqmrc2018

3. The first 18 Name solution ：https://github.com/PanXiebit/aic_rc

4. Testa score 73.2: https://github.com/antdlx/aic18_rc

5. ai challenger 2018 's final code: https://github.com/NoneWait/ai_challenge_2018_mrc

6. be based on capsule Viewpoint based reading comprehension model : https://github.com/freefuiiismyname/capsule-mrc

7. AI Challenger 2018 Read and understand the track code sharing ：https://github.com/renjunxiang/oqmrc_2018

8. Singularity wit can be shared in Testa Go beyond the first BERT programme ：BERT fine-tune Ultimate practice course

9. RCZoo: from Testa Great first 22 Name to Testb Great first 2 name , The author uses “[email protected]https://github.com/lixinsu/RCZoo” As a team name , however RCZoo It is more like a general deep learning reading comprehension 、 Q & a system solutions and toolbox , As for the details of the event, the author needs to reveal ：
Question answering, reading comprehension toolkit：https://github.com/lixinsu/RCZoo

English Chinese text machine translation

English Chinese machine text translation is one of the tasks of this competition , The goal is to evaluate the machine translation ability of each team . The machine translation language direction is English to Chinese . The test text is oral domain data . The participating teams need to train the machine translation system according to the data provided by the evaluator , You can freely choose machine translation technology . for example , Rule based translation technology 、 Statistical machine translation and neural network machine translation . This competition will use the objective evaluation index of machine translation （BLEU、NIST score、TER） Score ,BLEU The score will be used as the main machine evaluation index . The organizing committee will adopt objective indicators , Combined with the performance of the defense , Algorithm model for comprehensively evaluating contestants .

For more information, please refer to the home page of the track ：https://challenger.ai/competition/ect2018

Relevant code or solution ：

1. official baseline: be based on tensor2tensor and transformer Solutions for , It is estimated that it is also the plan of most players in this track , The difference lies in the handling of details and parameters ：
https://github.com/AIChallenger/AI_Challenger_2018/tree/master/Baselines/english_chinese_machine_translation_baseline

2. TestB The first 10 Share with classmates ：AI Challenger_2018 English Chinese text machine translation _ Competition summary
AI Challenger_2018 English Chinese text machine translation _ Competition summary - You know

3. Document-Transformer
This English Chinese machine translation competition provides context （Document) corpus , I haven't found the right solution , Until I saw Tsinghua 、 Sogou and Suzhou University recently produced this paper and Toolkit , Maybe this is the key to this competition , Interested students can have a try ：

Code ：https://github.com/Glaceon31/Document-Transformer
The paper ：Improving the Transformer Translation Model with Document-Level Context