package eu.faircode.email;
/*
This file is part of FairEmail.
FairEmail is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
FairEmail is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with FairEmail. If not, see .
Copyright 2018-2021 by Marcel Bokhorst (M66B)
*/
import android.content.Context;
import android.content.SharedPreferences;
import android.os.Build;
import android.text.TextUtils;
import androidx.preference.PreferenceManager;
import org.jetbrains.annotations.NotNull;
import org.json.JSONArray;
import org.json.JSONException;
import org.json.JSONObject;
import java.io.File;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.Comparator;
import java.util.Date;
import java.util.HashMap;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import javax.mail.Address;
import javax.mail.internet.InternetAddress;
public class MessageClassifier {
private static boolean loaded = false;
private static boolean dirty = false;
private static final Map> classMessages = new HashMap<>();
private static final Map>> wordClassFrequency = new HashMap<>();
private static final double CHANCE_MINIMUM = 0.20;
private static final double CHANCE_THRESHOLD = 2.0;
static void classify(EntityMessage message, EntityFolder folder, EntityFolder target, Context context) {
try {
if (!isEnabled(context))
return;
if (!canClassify(folder.type))
return;
if (target != null && !canClassify(target.type))
return;
File file = message.getFile(context);
if (!file.exists())
return;
long start = new Date().getTime();
// Build text to classify
StringBuilder sb = new StringBuilder();
List addresses = new ArrayList<>();
if (message.from != null)
addresses.addAll(Arrays.asList(message.from));
if (message.to != null)
addresses.addAll(Arrays.asList(message.to));
if (message.cc != null)
addresses.addAll(Arrays.asList(message.cc));
if (message.bcc != null)
addresses.addAll(Arrays.asList(message.bcc));
if (message.reply != null)
addresses.addAll(Arrays.asList(message.reply));
for (Address address : addresses) {
String email = ((InternetAddress) address).getAddress();
String name = ((InternetAddress) address).getAddress();
if (!TextUtils.isEmpty(email)) {
sb.append(email).append('\n');
int at = email.indexOf('@');
String domain = (at < 0 ? null : email.substring(at + 1));
if (!TextUtils.isEmpty(domain))
sb.append(domain).append('\n');
}
if (!TextUtils.isEmpty(name))
sb.append(name).append('\n');
}
if (message.subject != null)
sb.append(message.subject).append('\n');
String text = HtmlHelper.getFullText(file);
sb.append(text);
if (sb.length() == 0)
return;
// Load data if needed
load(context);
// Initialize data if needed
if (!classMessages.containsKey(folder.account))
classMessages.put(folder.account, new HashMap<>());
if (!wordClassFrequency.containsKey(folder.account))
wordClassFrequency.put(folder.account, new HashMap<>());
// Classify text
String classified = classify(folder.account, folder.name, sb.toString(), target == null, context);
long elapsed = new Date().getTime() - start;
EntityLog.log(context, "Classifier" +
" folder=" + folder.name +
" message=" + message.id +
"@" + new Date(message.received) +
":" + message.subject +
" class=" + classified +
" re=" + message.auto_classified +
" elapsed=" + elapsed);
Integer m = classMessages.get(folder.account).get(folder.name);
m = (m == null ? 0 : m) + (target == null ? 1 : -1);
if (m <= 0)
classMessages.get(folder.account).remove(folder.name);
else
classMessages.get(folder.account).put(folder.name, m);
Log.i("Classifier " + folder.name + "=" + m + " msgs");
dirty = true;
// Auto classify
if (classified != null &&
!classified.equals(folder.name) &&
!message.auto_classified &&
!EntityFolder.JUNK.equals(folder.type)) {
DB db = DB.getInstance(context);
try {
db.beginTransaction();
EntityFolder dest = db.folder().getFolderByName(folder.account, classified);
if (dest != null && dest.auto_classify) {
EntityOperation.queue(context, message, EntityOperation.MOVE, dest.id, false, true);
message.ui_hide = true;
}
db.setTransactionSuccessful();
} finally {
db.endTransaction();
}
}
} catch (Throwable ex) {
Log.e(ex);
}
}
private static String classify(long account, String currentClass, String text, boolean added, Context context) {
int maxMessages = 0;
for (String clazz : classMessages.get(account).keySet()) {
int count = classMessages.get(account).get(clazz);
if (count > maxMessages)
maxMessages = count;
}
State state = new State();
// First word
process(account, currentClass, added, null, state);
// Process words
if (Build.VERSION.SDK_INT < Build.VERSION_CODES.N) {
java.text.BreakIterator boundary = java.text.BreakIterator.getWordInstance();
boundary.setText(text);
int start = boundary.first();
for (int end = boundary.next(); end != java.text.BreakIterator.DONE; end = boundary.next()) {
String word = text.substring(start, end);
process(account, currentClass, added, word, state);
start = end;
}
} else {
// The ICU break iterator works better for Chinese texts
android.icu.text.BreakIterator boundary = android.icu.text.BreakIterator.getWordInstance();
boundary.setText(text);
int start = boundary.first();
for (int end = boundary.next(); end != android.icu.text.BreakIterator.DONE; end = boundary.next()) {
String word = text.substring(start, end);
process(account, currentClass, added, word, state);
start = end;
}
}
// Last word
process(account, currentClass, added, null, state);
if (!added)
return null;
if (maxMessages == 0) {
Log.i("Classifier no messages account=" + account);
return null;
}
// Calculate chance per class
DB db = DB.getInstance(context);
List chances = new ArrayList<>();
for (String clazz : state.classStats.keySet()) {
EntityFolder folder = db.folder().getFolderByName(account, clazz);
if (folder == null) {
Log.w("Classifier no folder class=" + account + ":" + clazz);
continue;
}
Stat stat = state.classStats.get(clazz);
double chance = stat.totalFrequency / maxMessages / state.words.size();
Chance c = new Chance(clazz, chance);
chances.add(c);
EntityLog.log(context, "Classifier " + c +
" frequency=" + (Math.round(stat.totalFrequency * 100.0) / 100.0) + "/" + maxMessages + " msgs" +
" matched=" + stat.matchedWords + "/" + state.words.size() + " words" +
" text=" + TextUtils.join(", ", stat.words));
}
if (BuildConfig.DEBUG)
Log.i("Classifier words=" + TextUtils.join(", ", state.words));
if (chances.size() <= 1)
return null;
// Sort classes by chance
Collections.sort(chances, new Comparator() {
@Override
public int compare(Chance c1, Chance c2) {
return -c1.chance.compareTo(c2.chance);
}
});
SharedPreferences prefs = PreferenceManager.getDefaultSharedPreferences(context);
double class_min_chance = prefs.getInt("class_min_chance", 20) / 100.0;
double class_min_difference = prefs.getInt("class_min_difference", 50) / 100.0;
// Select best class
String classification = null;
double c0 = chances.get(0).chance;
double c1 = chances.get(1).chance;
if (c0 > class_min_chance && c1 < c0 * class_min_difference)
classification = chances.get(0).clazz;
Log.i("Classifier current=" + currentClass +
" c0=" + Math.round(c0 * 100 * 100) / 100.0 + ">" + Math.round(class_min_chance * 100 * 100) / 100.0 + "%" +
" c1=" + Math.round(c1 * 100 * 100) / 100.0 + "<" + Math.round(c0 * class_min_difference * 100 * 100) / 100.0 + "%" +
" (" + class_min_difference + "%)" +
" classified=" + classification);
return classification;
}
private static void process(long account, String currentClass, boolean added, String word, State state) {
if (word != null) {
word = word.trim().toLowerCase();
if (word.length() < 2 ||
state.words.contains(word) ||
word.matches(".*\\d.*"))
return;
}
state.words.add(word);
if (state.words.size() < 3)
return;
String before = state.words.get(state.words.size() - 3);
String current = state.words.get(state.words.size() - 2);
String after = state.words.get(state.words.size() - 1);
Map classFrequency = wordClassFrequency.get(account).get(current);
if (added) {
if (classFrequency == null) {
classFrequency = new HashMap<>();
wordClassFrequency.get(account).put(current, classFrequency);
}
for (String clazz : classFrequency.keySet()) {
Frequency frequency = classFrequency.get(clazz);
if (frequency.count > 0) {
Stat stat = state.classStats.get(clazz);
if (stat == null) {
stat = new Stat();
state.classStats.put(clazz, stat);
}
int c = frequency.count;
Integer b = (before == null ? null : frequency.before.get(before));
Integer a = (after == null ? null : frequency.after.get(after));
double f = ((b == null ? 0 : b) + c + (a == null ? 0 : a)) / 3.0;
stat.totalFrequency += f;
stat.matchedWords++;
if (stat.matchedWords > state.maxMatchedWords)
state.maxMatchedWords = stat.matchedWords;
if (BuildConfig.DEBUG)
stat.words.add(current);
}
}
Frequency c = classFrequency.get(currentClass);
if (c == null)
c = new Frequency();
c.add(before, after, 1);
classFrequency.put(currentClass, c);
} else {
Frequency c = (classFrequency == null ? null : classFrequency.get(currentClass));
if (c != null)
c.add(before, after, -1);
}
}
static synchronized void save(Context context) throws JSONException, IOException {
if (!dirty)
return;
File file = getFile(context);
Helper.writeText(file, toJson().toString(2));
dirty = false;
Log.i("Classifier data saved");
}
private static synchronized void load(Context context) throws IOException, JSONException {
if (loaded || dirty)
return;
wordClassFrequency.clear();
File file = getFile(context);
if (file.exists()) {
String json = Helper.readText(file);
fromJson(new JSONObject(json));
}
loaded = true;
Log.i("Classifier data loaded");
}
static synchronized void clear(Context context) {
wordClassFrequency.clear();
dirty = true;
Log.i("Classifier data cleared");
}
static boolean isEnabled(Context context) {
SharedPreferences prefs = PreferenceManager.getDefaultSharedPreferences(context);
return prefs.getBoolean("classification", false);
}
static boolean canClassify(String folderType) {
return EntityFolder.INBOX.equals(folderType) ||
EntityFolder.JUNK.equals(folderType) ||
EntityFolder.USER.equals(folderType);
}
static File getFile(Context context) {
return new File(context.getFilesDir(), "classifier.json");
}
static JSONObject toJson() throws JSONException {
JSONArray jmessages = new JSONArray();
for (Long account : classMessages.keySet())
for (String clazz : classMessages.get(account).keySet()) {
JSONObject jmessage = new JSONObject();
jmessage.put("account", account);
jmessage.put("class", clazz);
jmessage.put("count", classMessages.get(account).get(clazz));
jmessages.put(jmessage);
}
JSONArray jwords = new JSONArray();
for (Long account : wordClassFrequency.keySet())
for (String word : wordClassFrequency.get(account).keySet()) {
Map classFrequency = wordClassFrequency.get(account).get(word);
for (String clazz : classFrequency.keySet()) {
Frequency f = classFrequency.get(clazz);
JSONObject jword = new JSONObject();
jword.put("account", account);
jword.put("word", word);
jword.put("class", clazz);
jword.put("frequency", f.count);
jword.put("before", from(f.before));
jword.put("after", from(f.after));
jwords.put(jword);
}
}
JSONObject jroot = new JSONObject();
jroot.put("version", 1);
jroot.put("messages", jmessages);
jroot.put("words", jwords);
return jroot;
}
private static JSONObject from(Map map) throws JSONException {
JSONObject jmap = new JSONObject();
for (String key : map.keySet())
jmap.put(key, map.get(key));
return jmap;
}
static void fromJson(JSONObject jroot) throws JSONException {
int version = jroot.optInt("version");
if (version < 1)
return;
JSONArray jmessages = jroot.getJSONArray("messages");
for (int m = 0; m < jmessages.length(); m++) {
JSONObject jmessage = (JSONObject) jmessages.get(m);
long account = jmessage.getLong("account");
if (!classMessages.containsKey(account))
classMessages.put(account, new HashMap<>());
String clazz = jmessage.getString("class");
int count = jmessage.getInt("count");
classMessages.get(account).put(clazz, count);
}
JSONArray jwords = jroot.getJSONArray("words");
for (int w = 0; w < jwords.length(); w++) {
JSONObject jword = (JSONObject) jwords.get(w);
long account = jword.getLong("account");
if (!wordClassFrequency.containsKey(account))
wordClassFrequency.put(account, new HashMap<>());
String word = jword.getString("word");
Map classFrequency = wordClassFrequency.get(account).get(word);
if (classFrequency == null) {
classFrequency = new HashMap<>();
wordClassFrequency.get(account).put(word, classFrequency);
}
Frequency f = new Frequency();
f.count = jword.getInt("frequency");
if (jword.has("before"))
f.before = from(jword.getJSONObject("before"));
if (jword.has("after"))
f.after = from(jword.getJSONObject("after"));
classFrequency.put(jword.getString("class"), f);
}
}
private static Map from(JSONObject jmap) throws JSONException {
Map result = new HashMap<>(jmap.length());
Iterator iterator = jmap.keys();
while (iterator.hasNext()) {
String key = iterator.next();
result.put(key, jmap.getInt(key));
}
return result;
}
private static class State {
private int maxMatchedWords = 0;
private List words = new ArrayList<>();
private Map classStats = new HashMap<>();
}
private static class Frequency {
private int count = 0;
private Map before = new HashMap<>();
private Map after = new HashMap<>();
private void add(String b, String a, int c) {
if (count + c < 0)
return;
count += c;
if (b != null) {
Integer x = before.get(b);
before.put(b, (x == null ? 0 : x) + c);
}
if (a != null) {
Integer x = after.get(a);
after.put(a, (x == null ? 0 : x) + c);
}
}
}
private static class Stat {
private int matchedWords = 0;
private double totalFrequency = 0;
private List words = new ArrayList<>();
}
private static class Chance {
private String clazz;
private Double chance;
private Chance(String clazz, Double chance) {
this.clazz = clazz;
this.chance = chance;
}
@NotNull
@Override
public String toString() {
return clazz + "=" + Math.round(chance * 100.0 * 100.0) / 100.0 + "%";
}
}
}