CTR　prediction　is　an　important　research　direction　in　the　field　of　recommendation　system.　In　recent　years,　DeepFM　has　been　concerned　because　it　can　learn　the　low-　and　high-order　feature　interactions　end-to-end,　and　it　has　become　one　of　the　best　models　for　CTR　prediction.　However,　due　to　the　inadequate　feature　interactions　in　the　FM　component　of　DeepFM　model,　the　prediction　accuracy　is　reduced.　Compared　with　DeepFFM,　although　the　FFM　component　of　DeepFFM　model　attributes　the　same　feature　to　the　same　filed　to　alleviates　the　inadequate　feature　interactions　in　FM　model,　its　binomial　parameter　number　is　far　greater　than　DeepFM,　which　increases　the　pressure　of　recommendation　system.　In　this　paper,　we　propose　a　new　model　called　DeepFaFM.　It　provides　a　scalable　coding　matrix　with　adjustable　factors,　which　can　dynamically　balance　the　operation　pressure　according　to　the　actual　production　situation.　Compared　with　DeepFM,　which　is　one　of　the　latest　classification　models,　it　has　the　advantages　of　DeepFM,　improves　the　inadequate　feature　interactions　in　the　operation　of　DeepFM,　and　improves　the　accuracy　of　the　model.　Experiments　show　that　DeepFaFM　is　better　than　DeepFM　in　accuracy,　and　is　equivalent　to　DeepFFM　but　its　operation　efficiency　is　significantly　higher　than　DeepFFM.　In　particular,　DeepFaFM　has　a　scalable　variable,　which　can　dynamically　balance　the　accuracy　and　calculation　pressure　according　to　the　actual　production　situation,　improving　the　flexibility　of　the　model.　©　2020　IEEE.