feat(neuro): split channel names into words and add Chinese topic words

app/src/androidTest/java/io/github/aedev/flow/data/recommendation/NeuroTextSegmentationTest.kt

@@ -40,6 +40,15 @@ class NeuroTextSegmentationTest {
     @Test
     fun thaiTitlesSplitIntoWords() = assertTopics("สอนเล่นกีตาร์สำหรับมือใหม่", "กีตาร์", "สำหรับ")
 
+    @Test
+    fun chineseChannelNamesGiveTheirTopicalWords() {
+        listOf("央視新聞" to "新聞", "央视新闻" to "新闻", "英文教室 科技" to "科技").forEach { (name, topic) ->
+            val topics = tokenizer.tokenizeChannelName(name)
+            assertTrue("$topic not in $topics for \"$name\"", topic in topics)
+        }
+        assertTrue(tokenizer.tokenizeChannelName("小明").isEmpty())
+    }
+
     @Test
     fun spacedScriptsAreUntouched() = assertTopics("Pixel 10 Pro camera test", "pixel", "camera")
 }

app/src/main/java/io/github/aedev/flow/data/recommendation/NeuroLanguages.kt

@@ -112,6 +112,23 @@ internal object NeuroLanguages {
         }
     }
 
+    private val TOPICAL_LANGUAGES = listOf("zh")
+
+    /**
+     * Words of channel names in scripts written without spaces that name a topic, folded
+     * (`resources/neuro/topical`). English has [NeuroTokenizer]'s own list. Read once per process.
+     */
+    val channelTopics: Set<String> by lazy {
+        buildSet {
+            TOPICAL_LANGUAGES.forEach { language ->
+                val stream = NeuroLanguages::class.java.getResourceAsStream("/neuro/topical/$language.txt") ?: return@forEach
+                stream.bufferedReader(Charsets.UTF_8).useLines { lines ->
+                    lines.filter { it.isNotBlank() && !it.startsWith("#") }.mapTo(this) { NeuroText.fold(it.trim()) }
+                }
+            }
+        }
+    }
+
     /**
      * The language of [pieces], the folded whitespace-separated pieces of a title, or null when
      * unsure. Scripts used by one language decide alone; Latin titles need two function words and a

app/src/main/java/io/github/aedev/flow/data/recommendation/NeuroTokenizer.kt

@@ -2935,21 +2935,17 @@ internal class NeuroTokenizer {
      * "Python Engineer"       → ["python"]      (engineer not in list, python is)
      * "Binging with Babish"   → []              (no topic words)
      * "History Matters"       → ["history"]
+     * "央視新聞"               → ["新聞"]        (央視 = a brand, 新聞 = news)
      */
     fun tokenizeChannelName(channelName: String): List<String> {
-        val rawOriginal =
-            NeuroText
-                .fold(channelName)
-                .split(WHITESPACE_REGEX)
-                .map { word -> word.trim { !it.isLetterOrDigit() } }
-                .filter { it.length > 1 }
+        val rawOriginal = nameWords(channelName)
 
         if (rawOriginal.isEmpty()) return emptyList()
 
         if (rawOriginal.size == 1) {
             val word = rawOriginal[0].lowercase()
             val lemma = keyOf(word)
-            return if (lemma in alwaysTopical || word in alwaysTopical) {
+            return if (isChannelTopic(lemma) || isChannelTopic(word)) {
                 listOf(lemma)
             } else {
                 emptyList()
@@ -2962,13 +2958,22 @@ internal class NeuroTokenizer {
             val lemma = keyOf(lower)
 
             when {
-                lemma in alwaysTopical -> topical.add(lemma)
-                lower in alwaysTopical -> topical.add(lower)
+                isChannelTopic(lemma) -> topical.add(lemma)
+                isChannelTopic(lower) -> topical.add(lower)
             }
         }
         return topical.distinct()
     }
 
+    private fun isChannelTopic(word: String): Boolean = word in alwaysTopical || word in NeuroLanguages.channelTopics
+
+    /** Words of a channel name; scripts written without spaces are split by the word breaker. */
+    private fun nameWords(name: String): List<String> =
+        NeuroText
+            .words(NeuroText.fold(name))
+            .map { word -> word.trim { !NeuroText.isWordChar(it) } }
+            .filter { it.length > 1 }
+
     /**
      * Builds a flat list of all meaningful words from the video's title,
      * channel name, and description. Used as context for domain disambiguation.
@@ -2986,12 +2991,7 @@ internal class NeuroTokenizer {
             .filter { it.length > 1 }
             .forEach { words.add(it) }
 
-        NeuroText
-            .fold(video.channelName)
-            .split(WHITESPACE_REGEX)
-            .map { it.trim { c -> !c.isLetterOrDigit() } }
-            .filter { it.length > 1 }
-            .forEach { words.add(it) }
+        words.addAll(nameWords(video.channelName))
 
         if (!video.description.isNullOrBlank()) {
             video.description

app/src/main/resources/neuro/topical/zh.txt

@@ -0,0 +1,110 @@
+# Channel-name topics for Chinese, Traditional and Simplified spellings. Each entry is one word of the platform word breaker.
+新聞
+新闻
+英文
+英語
+英语
+日语
+韓語
+韩语
+科技
+遊戲
+游戏
+音樂
+音乐
+電影
+电影
+美食
+料理
+旅遊
+旅游
+旅行
+健身
+財經
+财经
+投資
+投资
+股票
+汽車
+汽车
+攝影
+摄影
+動畫
+动画
+籃球
+篮球
+棒球
+足球
+網球
+网球
+數學
+数学
+物理
+化學
+化学
+生物
+歷史
+历史
+地理
+經濟
+经济
+政治
+哲學
+哲学
+心理
+心理學
+紀錄片
+喜劇
+喜剧
+手機
+手机
+烹飪
+烹饪
+烘焙
+寵物
+宠物
+科學
+科学
+教育
+語言
+语言
+健康
+瑜伽
+運動
+运动
+時尚
+时尚
+鋼琴
+钢琴
+吉他
+小提琴
+爵士
+搖滾
+摇滚
+嘻哈
+古典
+天文
+太空
+宇宙
+駭客
+電腦
+电脑
+相機
+相机
+機車
+摩托車
+腳踏車
+自行車
+裝修
+装修
+木工
+手工
+刺繡
+刺绣
+編織
+编织
+創業
+创业
+行銷
+商業
+商业

app/src/test/java/io/github/aedev/flow/data/recommendation/NeuroChannelNameTest.kt

@@ -0,0 +1,48 @@
+/*
+ * Copyright (C) 2025-2026 Flow | A-EDev
+ *
+ * This file is part of Flow (https://github.com/Flow-Tube/Flow).
+ */
+
+package io.github.aedev.flow.data.recommendation
+
+import com.google.common.truth.Truth.assertThat
+import com.google.common.truth.Truth.assertWithMessage
+import org.junit.Test
+
+/** Channel names give topics only for words the engine knows (#1341). */
+class NeuroChannelNameTest {
+    private val tokenizer = NeuroTokenizer()
+
+    @Test
+    fun `spaced names keep their topical words`() {
+        assertThat(tokenizer.tokenizeChannelName("Linus Tech Tips")).containsExactly("tech")
+        assertThat(tokenizer.tokenizeChannelName("Python Engineer")).containsExactly("python")
+        assertThat(tokenizer.tokenizeChannelName("Fireship")).isEmpty()
+        assertThat(tokenizer.tokenizeChannelName("Binging with Babish")).isEmpty()
+    }
+
+    @Test
+    fun `a name of a script without spaces has no topics unless it holds a topical word`() {
+        assertThat(tokenizer.tokenizeChannelName("")).isEmpty()
+        assertThat(tokenizer.tokenizeChannelName("央")).isEmpty()
+    }
+
+    @Test
+    fun `the Chinese topic list is folded, whole words`() {
+        val topics = NeuroLanguages.channelTopics
+        assertThat(topics).contains("新聞")
+        assertThat(topics).contains("新闻")
+        topics.forEach { word ->
+            assertWithMessage(word).that(word).isEqualTo(NeuroText.fold(word))
+            assertWithMessage(word).that(word.length).isAtLeast(2)
+            assertWithMessage(word).that(word.any(Char::isWhitespace)).isFalse()
+        }
+    }
+
+    @Test
+    fun `no topic word is also a stop word`() {
+        val stops = NeuroLanguages.stopList("zh")
+        assertThat(NeuroLanguages.channelTopics.filter { it in stops }).isEmpty()
+    }
+}