Of Men, Women, and Computers: Data-Driven Gender Modeling for Improved User Interfaces Hugo Liu Massachusetts Institute of Technology hugo@media.mit.edu Rada Mihalcea University of North Texas rada@cs.unt.edu Abstract Men and women have unique sensibilities for information, which can be tapped to create gender-sensitive user interfaces that appeal more specifically to each sex. Building on previous research in gen- der psychology and also in user modeling, we take a data-driven ap- proach to understanding gender preferences by mining a large corpus of 150,000 weblog entries– half authored by men, half by women. This paper reports two kinds of contributions. First, we employ auto- matic language processing, semantic analysis, and reflexive ethnog- raphy to articulate gender preferences for several dimensions of gen- der space will provide valuable insight to user interface designers– time, color, size, socialness, affect, and cravings. Second, we employ statistical gender models to build GENDERLENS–a novel intelligent news filtering system that customizes news based on the gender of its reader. A user evaluation found that GENDERLENS successfully predicted men and women’s preferences for news, with statistical significance for four out of five news genres tested. Keywords gender analysis, natural language processing, user interfaces 1. Introduction Men are from Mars, women are from Venus –or so the genre of self- help relationship literature would have us believe. But there is truth even in this folk idea– in certain respects, men and women think and feel differently, and perceive, value and understand the world in their own ways. The philosophical inquiry of feminist epistemology that was begun at the turn of the 20th century sought to reveal the fun- damentally different modi operandi of men and women, concluding that much of the societal and cultural knowledge that is considered to be universal and generic actually assumes patriarchal priorities and values. The redoubtable feminine theorist Virginia Woolf de- constructed these ’universal’ values: “But it is obvious that the values of women differ very often from the values which have been made by the other sex; naturally, this is so. Yet it is the masculine values that prevail. Speaking crudely, football and sport are ’important’; the worship of fashion, the buying of clothes ’trivial’. And these values are inevitably trans- ferred from life to fiction. This is an important book, the critic assumes, because it deals with war. This is an insignificant book because it deals with the feelings of women in a drawing-room.” 1 1 Virginia Woolf: 1929/1989, A room of one’s own, Harvest Books, 73-74. ICWSM’2007 Boulder, Colorado, USA If one could uncover men and women’s fundamentally different pri- orities and values, then it may be possible to improve communica- tion between the genders, and to translate news and information into the language of each gender so as to appeal to their unique sensi- bilities. So what are the priorities and values of men and women? How does one go about uncovering these gender differences? To perform ’gender modeling’, we turn to the personal writings of men and women, accessing 150,000 textual entries pulled from the we- blogs authored by men and women. The sort of writing found in a weblog may be ideally suited to what we wish to discover, since we- blogs often give an intimate account of personal everyday life, and personal viewpoint unto current events. More than just language and syntax, weblogs contain ample evidence of experiences and percep- tions, which we attempted to uncover using corpus-based modeling and semantic analysis. Finally, there is nary any pressure for we- blog writings to be formal or patriarchal– often they are non-linear and cyclical, and quite free to express the sort of candid values that feminists advocated for in their own ´ ecriture f´ eminine movement. In this paper, we first describe a corpus-based approach to gender modeling in the context of related work in user modeling and gender psychology. Second, we perform a statistical analysis and ethno- graphic study of gender difference in order to bring to light some insights that are likely to pique the interests of humanists and user interface designers alike. Third, we present GENDERLENS, a novel system for news filtering based on gender models, and present a user study of the GENDERLENS. 2. Related work Field work in social and gender psychology has had much to say about the differences of men and women. The masculine is stereo- typed as detached, rational, and aggressive, and the feminine as nur- turing, gentle, and tactful [4]. While some stereotypes are unfounded, sociolinguists do affirm that some communication styles are gen- dered. It has been found that men and women differ on private ver- sus public speaking, on ”report talk” versus ”rapport talk”–these and other facets of relational dialectics are gendered and constitute so- called ”GenderLects” [25]. Given the reality of distinctions, should not intelligent user interfaces accommodate men and women’s dif- ferent requirements and preferences for how information is commu- nicated to them? The role of gender differences in interfaces has a small but growing body of research. In a study on gender differences in email preferences, [6] argued that ”men tend to focus discourse on hierarchy and independence, while women focus on intimacy and solidarity.” In investigating gender preference in multimedia inter- faces [21], researchers found that girls emphasized ”writing, colors, drawing, help” while boys emphasized ”control over the computer, sharp moves and many movements on the screen.” The present research also extends recent ethnographic studies ad- dressing the role of gendered language and the ”gender gap” in the blogosphere [12], the significance of gender differences in self-dis-