1// Copyright (c) 2011 The Chromium Authors. All rights reserved. 2// Use of this source code is governed by a BSD-style license that can be 3// found in the LICENSE file. 4// 5// This proto represents a machine learning model which is used to compute 6// the probability that a particular page visited by Chrome is phishing. 7// 8// Note: sine the machine learning model is trained on the server-side and then 9// downloaded onto the client it is important that this proto file stays in 10// sync with the server-side copy. Otherwise, the client may not be able to 11// parse the server generated model anymore. If you want to change this 12// protocol definition or you have questions regarding its format please contact 13// chrome-anti-phishing@googlegroups.com. 14 15syntax = "proto2"; 16 17option optimize_for = LITE_RUNTIME; 18 19package safe_browsing; 20 21// This protocol buffer represents a machine learning model that is used in 22// client-side phishing detection (in Chrome). The client extracts a set 23// of features from every website the user visits. Extracted features map 24// feature names to floating point values (e.g., PageSecureLinksFreq -> 0.9). 25// 26// To compute the phishing score (i.e., the probability that the website is 27// phishing) a scorer will simply compute the sum of all rule scores for a 28// given set of extracted features. The score of a particular rule corresponds 29// to the product of all feature values that are part of the rule times the 30// rule weight. If a feature has no value (i.e., is not part of the extracted 31// features) its value will be set to zero. The overall score is computed 32// by summing up all the rule scores. This overall score is a logodds and can 33// be converted to a probability like this: 34// p = exp(logodds) / (exp(logodds) + 1). 35// 36// To make it harder for phishers to reverse engineer our machine learning model 37// all the features in the model are hashed with a sha256 hash function. The 38// feature extractors also hash the extracted features before scoring happens. 39message ClientSideModel { 40 // In order to save some space we store all the hashed strings in a 41 // single repeated field and then the rules as well as page terms 42 // and page words refer to an index in that repeated field. All 43 // hashes are sha256 hashes stored in binary format. 44 repeated bytes hashes = 1; 45 46 message Rule { 47 // List of indexes into hashes above which are basically hashed 48 // features that form the current rule. 49 repeated int32 feature = 1; 50 51 // The weight for this particular rule. 52 required float weight = 2; 53 } 54 55 // List of rules which make up the model 56 repeated Rule rule = 2; 57 58 // List of indexes that point to the hashed page terms that appear in 59 // the model. The hashes are computed over page terms that are encoded 60 // as lowercase UTF-8 strings. 61 repeated int32 page_term = 3; 62 63 // List of hashed page words. The page words correspond to all words that 64 // appear in page terms. If the term "one two" is in the list of page terms 65 // then "one" and "two" will be in the list of page words. For page words 66 // we don't use SHA256 because it is too expensive. We use MurmurHash3 67 // instead. See: http://code.google.com/p/smhasher. 68 repeated fixed32 page_word = 4; 69 70 // Page terms in page_term contain at most this many page words. 71 required int32 max_words_per_term = 5; 72 73 // Model version number. Every model that we train should have a different 74 // version number and it should always be larger than the previous model 75 // version. 76 optional int32 version = 6; 77 78 // List of known bad IP subnets. 79 message IPSubnet { 80 // The subnet prefix is a valid 16-byte IPv6 address (in network order) that 81 // is hashed using sha256. 82 required bytes prefix = 1; 83 84 // Network prefix size in bits. Default is an exact-host match. 85 optional int32 size = 2 [default = 128]; 86 }; 87 repeated IPSubnet bad_subnet = 7; 88 89 // Murmur hash seed that was used to hash the page words. 90 optional fixed32 murmur_hash_seed = 8; 91 92 // Maximum number of unique shingle hashes per page. 93 optional int32 max_shingles_per_page = 9 [default = 200]; 94 95 // The number of words in a shingle. 96 optional int32 shingle_size = 10 [default = 4]; 97} 98