With the proliferation of social media data, Multimodal Named Entity Recognition (MNER) has received much attention; using different data modalities is crucial for the development of natural language processing and neural networks. However, existing methods suffer from two drawbacks: 1) text-image pairs in the data only sometimes correspond to each other, and it is impossible to rely on contextual information due to the short text nature of social media. 2) Despite the introduction of visual information, heterogeneity gaps may occur in previous complex fusion methods, leading to misidentification. This paper proposes a new synthetic image with a selected graphic alignment network(SAMNER) to address these challenges and construct a matching relationship between external images and text. To solve the graphic mismatch problem, we use a stable diffusion model to generate the images and perform entity labeling. Specifically, we generate images and perform entity labeling through the stable diffusion model to generate the image with the highest match to the text, filter the generated images by the internal image set to generate the best image, and then perform multimodal fusion to predict the entity labeling, we design a simple and effective multimodal attentional alignment mechanism to obtain a better visual representation, and we conduct a large number of experiments. The experiments prove that our model produces competitive results on the two publicly available datasets.

This paper revisits a long-standing question of the validity of the First-Order Approach (FOA) for Principal-Agent problems with moral hazard in the agent's hidden effort. The canonical formulation of the problem [Holmström, 1979] presents a technical challenge: the agent faces infinitely many incentive constraints. The standard approach has been to relax this set of constraints by replacing them with a local incentive constraint that requires the first-order derivative of the agent's utility at the chosen action to be zero. However, this relaxation results in a loss of optimality in general.

This paper studies a novel social commerce practice known as "help-and-haggle," whereby an online consumer can ask friends to help her "haggle" over the price of a product. Each time a friend agrees to help, the price is cut by a random amount, and if the consumer cuts the product price down to zero within a time limit, she will get the product for free; otherwise, the product reverts to the original price. Help-and-haggle enables the firm to promote its product and boost its social reach as consumers effectively refer their friends to the firm. We model the consumer's dynamic referral behavior in help-and-haggle and solve for the optimal price-cut distribution for the firm that trades off social reach, promotion expense, and product sales. Our results are as follows.

In this paper, the machine learning method is applied to predict the two-impulse lunar transfer velocity and orbital transfer window in a fast and efficient way. Firstly, the two-impulsive lunar transfer problem is described, and a trajectory design method is proposed to optimize the transfer based on Sequential Quadratic Programming algorithm. Then the geocentric relationship of the parking Low Earth Orbit (LEO) and the Moon is analyzed and the Earth-Centered Moon-to-Earth Plane Coordinate is introduced to transfer the inertial orbital elements into the pseudo ones. In this coordinate frame, domain knowledge suggests that the pseudo orbital elements can also serve as good learning features for the machine learning model. After generating the dataset, two machine learning models are established to estimate the orbital transfer window and orbital transfer velocity impulse respectively. Numerical simulations demonstrate that the machine learning models are efficient to estimate hundreds of transfer trajectories within 0.1 second, which shows much better performance than traditional orbit optimization method. The performance of two different machine learning algorithms is also assessed, where the neural network outperforms the gradient boosting model with the velocity impulse error less than 1m/s. It is also verified that the feature selection of pseudo orbital elements is appropriate which has smaller learning errors than that using the inertial orbital elements.