In the reward-based crowdfunding model, the main mechanism for financiers to obtain funds is to obtain financial support from investors through pre-determined reward commitments. Reward commitment options consist of an ordered set of rewards ranging from the lowest to the highest contribution amount for potential investors to freely choose from. To avoid selection bias, the expected utility of each reward commitment option is related to the amount of money invested, and high rewards are usually accompanied by high investment costs. For the financier, the investor’s selective preference for higher rewards can lead to more funds. If financiers know the relationship between reward commitments and group investment propensities, they can place the specified reward commitments in the most attractive position for investors and improve project financing performance. Studies have been conducted to explore the economic exchange relationship between individual investors and reward commitments from an individual perspective; whereas, as a group participation model, there are fewer discussions on group decision-making tendencies, which makes it difficult to provide an effective guide to reward design. Based on expectancy theory, this paper introduces the concept of centroid in engineering to quantify the level of group participation willingness, and constructs an econometric model with decision centroid as the dependent variable to explore the influence mechanism of the reward commitment types of crowdfunding projects on investors’ decision-making. Meanwhile, the text mining model combining LDA topic model and SBERT is employed to overcome the shortcomings of the single LDA topic model that lacks semantic correlation, and to classify the types of reward commitments into utilitarian value type and emotional value type, the former focuses on utility and functionality, including entity type and virtual type; while the latter pays attention to the user experience, including service type and participation type. The LDA topic model is used to identify the topic vectors of the reward description text, and topic coherence is used to determine the optimal number of topics. Next, the SBERT model is used to obtain the sentence vectors, and the fixed-size sentence vectors are outputted by Mean-pooling. Then, the t-SNE algorithm is used to connect the vectors, realize the low-dimension representation. Finally, text clustering is performed using K-means clustering algorithm to assign semantically and thematically similar words to the same class cluster. This paper takes Kickstarter, a representative reward-based crowdfunding platform, as the research object, and collects 9,166 projects and their 41,496 reward commitment options as the research data, and analyzes the impact using the STATA 18.0. The results show the number of reward commitment options and the number of reward commitment types have a positive effect on investors’ decision centroid. In terms of reward utility, offering utilitarian value rewards is more useful than emotional value rewards, suggesting that financiers should offer reward commitment types whose reward value more closely matches the monetary value offered by the investor. In terms of the fine-grained perspective of reward utility, entity rewards have the strongest positive effect and service rewards have the strongest negative effect, suggesting that entity rewards are more helpful in building investor trust in the financier’s ability to fulfill commitments. Financier’s nationality and reward description text detail affect investor expectancy and instrumentality, respectively. Both have a positive moderating role between reward commitments and decision centroid. Nationality enhances the positive effects of the number of reward commitment options, and utilitarian value rewards on decision centroid. Reward description text length enhances the positive effects of the number of reward commitment options, the number of reward commitment types, and utilitarian value rewards on decision centroid, and weakens the negative effects of service rewards.