Data　mining　of　students＇　scores　can　optimize　teaching　strategies　and　improve　the　teaching　quality　continuously.　However,　there　are　two　problems　in　the　current　researches　of　teaching　data　mining.　Firstly,　most　researchers　focus　on　students＇　performance　among　multiple　courses,　but　these　data　are　relatively　rough,　and　with　little　help　for　improving　specific　teaching　strategies　(e.g.,　using　different　teaching　and　assessment　methods)　of　a　single　course.　Secondly,　students＇　performance　not　only　depends　on　their　efforts,　but　also　is　affected　by　the　effectiveness　of　the　assessment　methods　made　by　teachers,　so　it　is　impossible　to　have　accurate　improvement　without　considering　factors　from　both　teacher　and　students＇　sides.　This　paper　takes　the　students＇　scores　of　computer　network　course　as　the　sample,　uses　k-means　to　cluster　students＇　scores　from　different　assessment　methods　based　on　fine　grained　teaching　contents,　in　order　to　get　more　accurate　teaching　improvement　strategies　later.　In　addition,　an　improved　Apriori　algorithm　is　proposed　with　the　quality　of　assessment　methods　taken　into　consideration,　for　association　rules　selection　and　teaching　strategies　improvement.　According　to　the　experiments,　our　method　can　take　into　account　the　diversity　of　course　assessment　methods　and　the　influence　of　both　teaching　and　learning　factors,　and　the　accuracy　of　the　improved　rules　is　18.8%　higher　than　that　current　Apriori　algorithm　based　on　interest.　©　2020　ACM.