At NetEase, the User Research Department is responsible for the user experience research work for most innovative Internet businesses. According to exclusive information obtained by the editor, NetEase's user research department is mainly divided into two lines: user research and strategic analysis . The user research line is responsible for the user research work of most of NetEase's innovative Internet businesses, and currently mainly serves more than ten Internet products such as music, education, social networking , smart hardware, and reading. Next is the time for the head teacher to knock on the blackboard User research methods can be divided into two main categories: qualitative research and quantitative research. Frequently used methods in qualitative research include: in-depth interviews, observation, usability testing, etc., while questionnaire surveys, traffic/log data analysis , experimental methods, etc. are mainly used in quantitative research. The validity of the research conclusions is guaranteed through a systematic methodology. What I want to talk about today is " user portrait ". The so-called user portrait is a character prototype of the target user. It can not only quickly understand the user's basic information and quickly classify it, but also further accurately analyze the user's behavioral habits and attitude preferences. Although user portraits are virtual representations of users, they must be based on real users and real data. 1Clarify the purpose of the study When we try to create a user portrait, it is often based on the following scenarios: 1. Determine the target users , divide the users into different types according to different characteristics, and determine the proportion and characteristics of the target users; 2. Collect user data to obtain user operation behavior, emotional preferences, demographic information, etc. 3. Determine product development priorities based on target users, and focus on target users’ motivations and behaviors in design and operation ; 4. Convenient design and operation . Product design and operation activities based on the specific character images provided by user portraits are more convenient and reliable than user images that are vague, fictitious, or based on personal preferences. 5. Build intelligent recommendation systems based on different types of users, such as personalized recommendations, precise operations, etc. It can also be seen from the usage scenarios of user portraits that user portraits are applicable to each product cycle: from potential user mining to new user attraction, to the cultivation of old users and the return of lost users, user portraits have their place. 2Clarify research methods User portraits can be constructed using qualitative methods (e.g., in-depth interviews, focus groups) or quantitative methods (e.g., quantitative questionnaires, behavioral log data). Different methods have their own advantages and disadvantages: However, whether you choose a qualitative or quantitative method, you first need to have a basic "quantitative" understanding of the user type, otherwise there will be bias in selecting samples. So how do we construct user portraits through quantitative methods (clustering)? 3Determine target dimensions and data3.1 Which indicators to choose? The selection of user indicators can be closed or open. In closed indicators , the type of user group is fixed, and all user types constitute the entire user group, such as light users, heavy users; male users, female users . However, this division method may be too single-dimensional and unable to reflect the complexity of the user group, and is not conducive to the supplementation, improvement and iteration of the indicator system. Therefore, in our research, we prefer to adopt an open classification method, which can change or expand indicators according to different application scenarios.The open indicator system includes user demographic attributes, behavioral operation attributes, attitude preference attributes, user value attributes, etc. User behavior and attitudes are constantly changing. Among them, it should be noted that the population attribute indicators in the closed indicators are relatively stable static data. Usually, from our experience and the user information we have, we have a clear idea of the user's age structure and gender ratio. If the demographic attribute indicators have a greater impact on the clustering (strong collinearity) or have too high an impact as a factor in the model, we can focus on indicators such as user behavior and attitude preferences when clustering. After successful clustering, we can compare the demographic background information of each user type. 3.2 How to obtain and filter data? After identifying the indicator, we need to determine the source of the indicator. Some data can be recorded in the background behavior log, and some requires questionnaire survey. Generally speaking, behavioral-level indicators can be more accurately measured using background logs. The attitude level should be obtained through questionnaires. In theory, all data can be obtained through questionnaires. However, in order to optimize the research results, we adopted a combination of questionnaire + behavior log . While sending the questionnaire, the user's device number and ID were captured to match the background data. On the premise of ensuring the validity of the questionnaire, questionnaire design also needs to pay attention to combining user characteristics to improve the response rate and data accuracy. For example, for 2D users, given that the age structure of the user group is relatively young, the questionnaire cannot be too long and cannot contain profound professional terms; at the same time, the sentence expression and page style of the questionnaire should also be adjusted accordingly to avoid a sense of distance. At the same time, be sure to screen out questionnaires that have been filled out multiple times and registered using fake accounts. In addition, attention should be paid to the proportion of new users. It is necessary to evaluate whether the proportion of newly registered users filling out the questionnaire is consistent with the normal increase in new users during the launch period. Whether the user portrait needs to include new users depends on the purpose of the project and can also be decided after discussion with the product side. 4Try and evaluate user clustering4.1 How many types of users can be divided into?Cluster analysis is an exploratory study. It determines the closeness of relationships based on the distance between indicators or variables and clusters similarities into one category . Therefore, there will be multiple possible solutions, and no optimal solution will be given. The final choice depends on the researcher's analysis and judgment. The fewer types of users are divided into, the coarser the granularity, and the characteristics of each type will not be very clear; the more user types there are, the finer the granularity, but complex type divisions will also bring burdens to product positioning and operational promotion . Therefore, refining the granularity requires not only quantitative clustering for adjustment, but also verification based on product experience. At the same time, because we use an open indicator system, it is impossible for us to know the number of user types as clearly as we distinguish between "male users and female users". Therefore, when using data to create user portraits, the most critical step is to determine how many types of users can be divided into. 4.2 How to choose a suitable clustering method? After determining the factors, you need to choose an appropriate clustering method. Different methods are applicable to different situations. The most commonly used ones are K-means clustering and hierarchical clustering.K-means clustering is also called fast clustering. It requires less memory, has low complexity, is fast and efficient, and is suitable for large amounts of data. However, the number of categories needs to be determined in advance and the mean needs to be defined. Only samples can be clustered, not variables, and the sample variables must be continuous variables.Hierarchical clustering can cluster variables or samples, and can be continuous variables or categorical variables. It can provide a variety of methods for calculating distance, but the calculation complexity is high and it is suitable for small data volumes. We need to determine the clustering method based on the specific circumstances of the project, including project cycle, data form, data volume, clustering characteristics, etc. Finally, by trying different numbers of clusters, distance algorithms, and classification methods, we can determine the number of classifications based on the following points: 1. Based on product experience, typical users of different products are different 2. Based on existing user research and related research conclusions 3. Determine based on specific classification results 4. According to the hierarchical clustering "steps-distance" inflection point The clustering effect can be evaluated from the distance between cluster centers, the variance within components and groups, whether the ratio between the number of groups conforms to the product characteristics, whether the ratio is coordinated, and whether the type of division is meaningful to the product. 5Restore data to user After knowing the classification results and analyzing the characteristics of each type of user in terms of various indicators, the work of building user portraits is like filling a skeleton with flesh and blood. On the one hand, we can directly use the acquired data to find information with significant features and assign it to users. For example, 60% of the first category of users use the iOS system, while the other three categories do not exceed 20%. We can abstract the first category of users as a person who usually uses an iPhone. In addition to questionnaire data, if you want to make the character image more vivid, you can analyze the questions in the questionnaire, or create a portrait based on product experience, user feedback or existing research, so that the user image can be more flesh and blood. However, restoring data to the user itself also needs to follow several principles. Persona means that a convincing user role must meet seven conditions: P stands for Primary research, which refers to whether the user persona is based on contextual interviews with real users.E stands for Empathy. It refers to whether the user persona, which includes a name, photo, and product-related description, is empathetic.R stands for Realistic and refers to whether the user persona seems like a real person to the people who interact with customers every day.S stands for Singular. Each user is unique and has few similarities with each other.O stands for Objectives. Does the user role contain high-level goals related to the product and does it contain keywords to describe the goals?N stands for Number. Is the number of personas small enough so that the design team can remember the name of each persona, as well as one primary persona?A stands for Applicable. Can the design team use the user persona as a practical tool to make design decisions? Note: The Persona principle comes from Alan Cooper, https://plus.google.com/101097598357299353681/about Through quantitative research, we can quickly establish an accurate understanding of users, conduct comparative statistical analysis on users of different numbers and characteristics, and in the later process of product iteration and improvement, we can prioritize users and focus on core, large-scale users. However, the user portrait established by relying on data in a relatively quantitative way is still rough, and it is difficult to describe the life situations and usage scenarios of typical users, and it is difficult to explore the reasons and deep motivations behind users' emotional tendencies and behavioral operations. Therefore, if you have enough energy and time, you can conduct in-depth interviews with each type of user and combine quantitative and qualitative methods to create a more accurate and vivid user portrait. After reading all the useful information above, I wonder if you have a better understanding of user portraits. At NetEase, user experience research work usually enters the project from the product concept design stage , helping the business side to understand the attitudes and behavioral characteristics of each user segment, and identify the core needs and business opportunities of users in the segmented market; in the product design and testing stage , user research is carried out to ensure that the product design is user-centric while identifying corresponding problems in usability and providing solutions, providing direction for product iteration and optimization; in the growth stage after the product is launched , analyze users' attitudes and evaluations on various aspects of the product, understand users' usage habits, and improve product satisfaction by continuously optimizing user experience, ensuring that the product finds and maintains a balance between commercialization and user experience.
Mobile application product promotion service: APP promotion service Qinggua Media information flow
The author of this article @Xiang YuCompiled and published by (APP Top Promotion), please indicate the author information and source when reprinting!
