v0.3.1 - Course Recommender Improvements

v0.3.1 - Course Recommender Improvements

This patch fixes three scoring logic bugs in the Course Recommender introduced in v0.3.0, improving recommendation accuracy and relevance by reducing irrelevant courses from appearing in results.

🐛 Bug Fixes

• Improved interest matching accuracy: Fixed an issue where metadata noise in memory files (e.g. timestamp fields) interfered with interest term matching, causing unrelated courses (finance, politics, language) to receive inflated scores.
• Improved completed course detection: Fixed an issue where course codes mentioned in the recommendation section of memory were incorrectly treated as already completed and penalized. Course codes are now only considered completed when in proximity to explicit completion keywords (e.g. completed / taken / 已修).
• Improved bilingual course preference inference: Fixed an issue where descriptive mentions of "双语" (bilingual) in memory were mistakenly inferred as a user language preference, awarding bonus scores to all bilingual courses. An explicit preference phrase (e.g. prefer / 偏好 / 可接受) is now required to trigger inference.

---

v0.3.1 - 课程推荐表现增强

本次更新针对 v0.3.0 课程推荐模块中三处评分逻辑进行修复，增强推荐结果的准确性与相关性，减少无关课程出现在推荐列表中的情况。

🐛 问题修复

• 增强课程推荐的兴趣匹配准确性：修复记忆文件中的元数据噪声（如时间戳字段）干扰兴趣词匹配的问题，避免金融、政治、语言等无关课程获得虚假高分。
• 增强已修课程的识别精度：修复推荐列表中提及的课程代码被误判为"已修课程"并遭降权排除的问题，现仅在明确出现完成关键词（如 completed / taken / 已修）时才视为已修。
• 增强双语课程偏好推断的准确性：修复记忆内容中对"双语"的描述性提及被误判为用户语言偏好、导致所有双语课程获得额外加分的问题，现要求明确的偏好表达短语方可触发推断。