User forgiveness behavior in the context of failed intelligent robot services is a key mechanism for the sustainability of human-machine relationships, but the existing research has not yet clarified its internal cognitive path and contextual boundaries. This study breaks through the limitations of traditional interpersonal forgiveness theory and constructs a chain model of “emotional commitment → responsibility attribution → forgiveness behavior” based on cognitive dissonance theory and attribution behavior theory, and explores the moderating effect of service failure types. Through a 2 × 2 inter group experimental design, it was found that emotional commitment positively influences forgiveness behavior through the partial mediating effect of external responsibility attribution, and service failure types significantly moderate the impact of emotional commitment on responsibility attribution. In social failure situations, users with high emotional commitment have a significantly increased tendency towards external attribution, while this effect is not significant in functional failure situations.