Vít Listík - Email.cz workshop
0 likes767 views
The document discusses email classification and summarization techniques used by Email.cz. It details their processes for analyzing spam sources like content, metadata, and blacklists. It also discusses challenges with "graymail" and techniques like natural language processing, support vector machines, image analysis using deep learning models, and decision trees for categorizing email topics in multiple languages. Distributed learning approaches are used to speed up image classification.
![NLP
● Bag of words
● Lemmatization
● Stop words
(1) John likes to watch movies. Mary likes movies too.
(2) John also likes to watch football games.
[
"John",
"likes",
"to",
"watch",
"movies",
"also",
"football",
"games",
"Mary",
"too"
]
(1) [1, 2, 1, 1, 2, 0, 0, 0, 1, 1]
(2) [1, 1, 1, 1, 0, 1, 1, 1, 0, 0]](https://image.slidesharecdn.com/listik-160525130337/85/Vit-Listik-Email-cz-workshop-15-320.jpg)
Vít Listík - Email.cz workshop
- 1. Email.cz workshop Vit Listik @tivwvit
- 2. Email stats ● 60M emails per day ● 3M users daily, 6M monthly ● 2 PB Email delivery process
- 3. Antispam ● Fighting with the bad guys
- 4. Antispam sources ● Content ○ Text ○ Images ○ Attachments ○ Links ○ Headers ● Metadata ○ Traffic ○ Historic data (reputation) ○ Blacklists ○ Rules (DKIM, DMARC, SPF)
- 7. Grey email Graymail is solicited bulk email messages that don't fit the definition of email spam (e.g., the recipient "opted into" receiving them). Recipient interest in this type of mailing tends to diminish over time, increasing the likelihood that recipients will report graymail as spam. In some cases, graymail can account for up to 82 percent of the average user's email inbox.
- 9. ML in antispam ● Topic ● Usubscribe ● Phishing ● Domain keywords ● Images ● Personalized filter ● Link naturalness
- 11. Tools ● Jupyter ○ Visualizations ○ State ● HDF5 ● Pandas ● Ipython cluster ● Cluster storage
- 12. Let's go (to) Jupyter
- 14. Topic categorization ● 16 categories ● Manually labeled dataset ● 2 languages (2 models) ● 7th version ● Overlapping classes
- 15. NLP ● Bag of words ● Lemmatization ● Stop words (1) John likes to watch movies. Mary likes movies too. (2) John also likes to watch football games. [ "John", "likes", "to", "watch", "movies", "also", "football", "games", "Mary", "too" ] (1) [1, 2, 1, 1, 2, 0, 0, 0, 1, 1] (2) [1, 1, 1, 1, 0, 1, 1, 1, 0, 0]
- 16. SVM ● Classification ● Best split for classes ● Linear classifier (kernels) Multi class version ● One vs. all ● Winner takes all
- 18. Image categorization ● Classes: spam x ham ● Based on user reaction ● Links analysis ● Low level image features ○ Size ○ DPI ○ Hists ○ Exif ○ Compression ● Raw pixels
- 19. Spam roulette
- 21. User reactions ● Noisy ● Inconsistent ● Bots ● Low ratio
- 22. Image topics ● Caffe ● Pretrained network ● Same classes as for words ● Cleaned dataset of images from classified emails ● 400k images ● Slow on CPU Loan non-bank Pharmacy DiscountEbola
- 23. Distributed learning ● Spark ● SparkNet (Caffe) ● Elepheas (Keras)
- 24. Image types ● Trivial ○ Animated ○ Monitoring ○ Border ● Photo ● Graphics ● Photo with graphics Graphics Photo
- 25. Image features Extraction ● PIL ● OpenCV ● Image Magick Features (142) ● Channel stats ○ Min, max, mean ○ Standard deviation ○ Skewness ○ Entropy
- 26. Learning ● Scipy - Decision Trees ● Keras (Tensorflow, theano) ● 30k Manually labeled samples
- 28. Message ● Gray email ● Explore (visualize) your data (in Jupyter) ● Use libraries ● Simple subtasks (boosting) may help ● Store intermediate results ● Store test results with the model