I290　scenes　features,　including　the　dependencies　between　the　attribute　information　and　user　ratings　are　implicit　variables　determine　a　user　score　of　commodities　to　build　hidden　variable　model　includes　user　preferences,　describe　any　form　of　dependence　between　the　rating　data　relevant　attributes　as　the　main　target,　The　Bayesian　network　is　used　as　the　basic　framework　of　the　dependencies　between　attributes　and　the　probability　relationship　between　attributes　and　scores.　The　project　scoring　data　is　used　to　construct　the　item　scoring　model　without　hidden　variables,　and　the　semi-clique　structure　is　proposed　to　insert　the　description　of　user　preferences　the　latent　variable　model　with　user　preference　is　constructed,　and　the　method　of　parameter　estimation　of　hidden　variable　model　based　on　EM　algorithm　is　given.　Then　the　probability　inference　model　of　the　hidden　variable　model　and　the　corresponding　project　score　prediction　method　are　proposed.　The　experimental　results　based　on　MovieLens　and　LDOS-CoMoDa　data　show　that　the　proposed　Bayesian　model　with　implicit　variables　and　the　corresponding　score　prediction　method　are　effective.　©　2018　IEEE.