#36: Update rules for better accuracy

These rules and coefficients yield the following accuracy based on a training corpus of 50 product pages from our top 5 sites (Amazon, Ebay, Walmart, Best Buy and Home Depot):
* 100% for product 'image'
* 96% for product 'title'
* 94% for product 'price'
Product 'price' and 'title' features have proximity rules based on the highest scoring product 'image' element. For now, this is done by accessing the image fnode using an internal '_ruleset' object; @erikrose is working on better support for this use case in the very near future, so this implementation can be improved at that time.

.gitignore

@@ -2,3 +2,4 @@ node_modules
 web-ext-artifacts
 build
 gecko.log
+.DS_Store

src/fathom_coefficients.json

@@ -1,9 +1,11 @@
 {
-  "largerImageCoeff": 3,
-  "largerFontSizeCoeff": 1,
-  "hasDollarSignCoeff": 3,
-  "hasTitleInIDCoeff": 10,
-  "hasTitleInClassNameCoeff": 5,
-  "isHiddenCoeff": -100,
-  "isHeaderElementCoeff": 10
+  "largerImageCoeff": 2,
+  "largerFontSizeCoeff": 7,
+  "hasDollarSignCoeff": 8,
+  "hasPriceInIDCoeff": 17,
+  "hasPriceInClassNameCoeff": 2,
+  "isAboveTheFoldPriceCoeff": 33,
+  "isAboveTheFoldImageCoeff": 13,
+  "isNearbyImageXAxisCoeff": 5,
+  "hasPriceishPatternCoeff": 15
 }

src/fathom_extraction.js

@@ -15,10 +15,12 @@ import {
   largerImageCoeff,
   largerFontSizeCoeff,
   hasDollarSignCoeff,
-  hasTitleInIDCoeff,
-  hasTitleInClassNameCoeff,
-  isHiddenCoeff,
-  isHeaderElementCoeff,
+  hasPriceInIDCoeff,
+  hasPriceInClassNameCoeff,
+  isAboveTheFoldPriceCoeff,
+  isAboveTheFoldImageCoeff,
+  isNearbyImageXAxisCoeff,
+  hasPriceishPatternCoeff,
 } from 'commerce/fathom_coefficients.json';
 
 const PRODUCT_FEATURES = ['title', 'price', 'image'];
@@ -36,10 +38,12 @@ function runRuleset(doc) {
       largerImageCoeff,
       largerFontSizeCoeff,
       hasDollarSignCoeff,
-      hasTitleInIDCoeff,
-      hasTitleInClassNameCoeff,
-      isHiddenCoeff,
-      isHeaderElementCoeff,
+      hasPriceInIDCoeff,
+      hasPriceInClassNameCoeff,
+      isAboveTheFoldPriceCoeff,
+      isAboveTheFoldImageCoeff,
+      isNearbyImageXAxisCoeff,
+      hasPriceishPatternCoeff,
     ]).against(doc).get(`${feature}`);
     fnodesList = fnodesList.filter(fnode => fnode.scoreFor(`${feature}ish`) >= SCORE_THRESHOLD);
     // It is possible for multiple elements to have the same highest score.