Multi-view　data　is　widely　used　in　the　real　world,　and　traditional　machine　learning　methods　are　not　specifically　designed　for　multi-view　data.　The　goal　of　multi-view　learning　is　to　learn　practical　patterns　from　the　divergent　data　sources.　However,　most　previous　researches　focused　on　fitting　feature　embedding　in　target　tasks,　so　researchers　put　forward　with　the　algorithm　which　aims　to　learn　appropriate　patterns　in　data　with　associative　properties.　In　this　paper,　a　multi-view　deep　matrix　factorization　model　is　proposed　for　feature　representation.　First,　the　model　constructs　a　multiple　input　neural　network　with　shared　hidden　layers　for　finding　a　low-dimensional　representation　of　all　views.　Second,　the　quality　of　representation　matrix　is　evaluated　using　discriminators　to　improve　the　feature　extraction　capability　of　matrix　factorization.　Finally,　the　effectiveness　of　the　proposed　method　is　verified　through　comparative　experiments　on　six　real-world　datasets.　©　2021,　Springer　Nature　Switzerland　AG.