For the multimodal sentiment analysis of commentary videos, different modalities contain target information with varying degrees of contribution. Existing methods tend to focus on mining textual modalities that contribute more, which inhibits modalities that contribute less during training, thereby weakening the effective information of modalities that contribute less during fusion. Hence, a modal-correlated generative adversarial network (MGAN) for the multimodal sentiment analysis of commentary videos is proposed, which balances modal contributions by differential complementation. First, temporal-based contextual features are extracted using long short-term memory to accommodate the characteristics of different modalities. The textual modality is identified as the primary contributing modality and the other two modalities as secondary contributing modalities. Second, an MGAN is proposed to generate complementary features with the same distribution as the secondary contributing modalities by correlating between the primary and secondary contributing modalities. To further mine information regarding the discrepancies between the primary and secondary contributing modalities, propose a semantic filtering strategy based on feature similarity. It obtains cross-modality-correlated contextual features by establishing a cross-modal contextual attention mechanism; based on these features, the cross-modal similarity is computed. Additionally, a gating mechanism is established to filter out weakly relevant semantics to enhance the proportion of secondary contributing modalities that differ from the primary contributing modalities and generate more purely differential complementary features that conform to sentiment polarity. Finally, multimodal semantic and differential complementary features are fused to fully utilize the discrepancies in target information contained among the modalities to complement the fusion-model information. Experimental results show that the proposed model performs better on multiple datasets, with an average increase of 3.45 in accuracy compared with the optimal baseline method, particularly on CMU-MOSEI, where the model achieves an accuracy level and F1 score of 92.5 % and 93.8 %, respectively. These results validate the effectiveness of proposed model for multiple datasets.

The incorporation of Laves phase into HEAs coating represents a novel approach to enhance wear resistance, building upon the innovative background of titanium alloy surface protection. The TiAlZrVNiX X (X = 1, 1.5, 2) HEAs coatings were prepared in this study using laser cladding advantages to achieve dependable wear-resistant coatings. The microstructure, phase composition, microhardness, and tribological properties of the HEA coatings were simultaneously investigated. TiAlZrVNiX X coatings were composed of V-rich HCP solid solution phases, C14- Laves phases, and NiTi2 2 phases. The microhardness of the 1.5Ni coating measured 708.75HV0.3 0.3 and wear rate of 1.12 x 10-- 13m3N- 1 m 3 N- 1 m- 1 . The primary wear mechanism was abrasive wear, and it had been confirmed that the 1.5Ni coating exhibited best wear resistance. The Laves phase with high space utilization would produce precipitation strengthening and dispersion strengthening, effectively reducing the stress-induced microcrack diffusion. Additionally, the introduction of Ni elements induced the growth of the NiTi2 2 phase in order to enhance the strength while mitigating brittleness, thereby enhancing the wear resistance of the coatings.

Graph neural networks (GNNs) have indeed shown significant potential in the field of personalized recommendation. The core approach is to reorganize interaction data into a user–item bipartite graph, leveraging high-order connectivity between user and item nodes to enhance their representations. However, most existing methods only deploy graph neural networks on parallel interaction graphs and treat the information propagated from all neighbors as equivalent, failing to adaptively capture user preferences. Therefore, the representations obtained may contain redundant or even noisy information, leading to non-robustness and suboptimal performance. The recently proposed Multimodal Graph Attention Network (MGAT) disentangles personal interests at the granularity of modality, operating on individual modal interaction graphs while utilizing a gated attention mechanism to differentiate the impacts of different modalities on user preferences. However, MGAT merely uses averaging to fuse multimodal features, which might overlook unique or critical information within each modality. To address this issue, this paper proposes a multimodal preference-based graph attention network. Firstly, for each individual modality, a single-modality graph network is constructed by integrating the user–item interaction bipartite graph, enabling it to learn user preferences for that modality. GNNs are used to aggregate neighborhood information and enhance the representation of each node. Additionally, a GRU module is utilized to determine whether to aggregate neighborhood information, thereby achieving noise reduction. In addition, a lightweight complementary attention mechanism is proposed to fuse user and item representations learned from different modal graphs. The complementary attention mechanism can not only avoid information redundancy, but also alleviate the problem of modal loss to a certain extent, and ultimately input the fusion results into the prediction module. Experimental results on the MovieLens and TikTok datasets demonstrate the effectiveness of the multimodal information and attention fusion mechanism in improving recommendation accuracy. Compared to baseline state-of-the-art algorithms, the proposed model achieves significant improvements in the Precision@K, Recall@K, NDCG@K and AUC metrics. Our code is publicly available on GitHub: [ https://github.com/Oasisway624/mgpat ].

Traditional Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) struggle with semantic comprehension and long-distance dependency capture in recommendation systems. To address this, we proposeMBTRec, a multimodal recommendation model based on theTransformer encoder. It employs an innovative bidirectional tower-type attention mechanism (Bi Towernet) for modal fusion, ensuring the independent contribution of each modality while optimizing interaction and feature representation. MBTRec integrates forgetting functions and StreamLDA techniques to capture users' dynamic interest topics and uses Deep Canonical Correlation Analysis (DCCA) to explore the correlation between topics and multimodal information. Through dense incremental dynamic time windows, MBTRec captures users' latest preferences and leverages the Transformer model to predict recommendation outcomes.

The serious wear damage caused by low surface hardness and poor wear resistance of titanium alloy severely limits its further application in engineering field. In this work, novel eutectic high entropy alloys (EHEAs) were designed by means of infinite solution strategy combined with mechanical assistance to explore the application of EHEAs in the field of surface protection coating. Three types of EHEA coatings were achieved through the adjustment of the Cr and Zr element proportions on Ti-6Al-4V alloy by laser cladding. The microstructure was eutectic structure composed of Laves and BCC/B2 phases with superlattice structure. A significant presence of stacking faults, Lomer-Cottrell locks and other defects is observed within the BCC/B2 phase, which contributes to the enhancement of both microhardness and nano-hardness. Al27Cr25Ti18Nb18Zr12 coating showed excellent wear resistance and minimal volume loss, which are 0.340 (65.38 % of the substrate) and 6.71 × 10−5 mm3/N·m (16.89 % of the substrate). Notably, the three alloys exhibit interesting selective corrosion behavior, which is mitigated by the presence of a nano-lamellar eutectic structure that limits pitting propagation.

This article has been retracted. A retraction notice can be found at https://doi.org/10.3233/JIFS-219433.

A novel Al27Cr25Ti18Nb18Zr12 eutectic high entropy alloy (EHEA) coating with BCC/B2 and Laves phase has been developed by laser cladding on Ti-6Al-4V alloys. The average microhardness of EHEA coating was ∼697.6HV0.3, which significantly surpasses the values reported for most eutectic high-entropy alloy coatings. The strengthening effect and additional strain hardening sources generated by the intersecting nanometer-spaced stacking faults and Lomer-Cotrell locks which found in BCC/B2 phase are important reasons for the superior hardness. And the adhesive problem of Ti-6Al-4V alloy was ameliorated by the EHEA coating with lower friction coefficient (83.05% of the Ti-6Al-4V alloy) and volumetric loss (27.75% of the Ti-6Al-4V alloy).

The serious wear damage caused by low surface hardness and poor wear resistance of titanium alloy severely limits its further application in engineering field. In this work, novel eutectic high entropy alloys (EHEAs) were designed by means of infinite solution strategy combined with mechanical assistance to explore the application of EHEAs in the field of surface protection coating. Three types of EHEA coatings were achieved through the adjustment of the Cr and Zr element proportions on Ti-6Al-4V alloy by laser cladding. The microstructure was eutectic structure composed of Laves and BCC/B2 phases with superlattice structure. A significant presence of stacking faults, Lomer-Cottrell locks and other defects is observed within the BCC/B2 phase, which contributes to the enhancement of both microhardness and nano-hardness. Al27Cr25Ti18Nb18Zr12 27 Cr 25 Ti 18 Nb 18 Zr 12 coating showed excellent wear resistance and minimal volume loss, which are 0.340 (65.38% of the substrate) and 6.71 x 10-5-5 mm3/N & sdot;m 3 /N & sdot;m (16.89 % of the substrate). Notably, the three alloys exhibit interesting selective corrosion behavior, which is mitigated by the presence of a nano-lamellar eutectic structure that limits pitting propagation.

In this work, a novel lightweight AlCrTiNbMo refractory high-entropy alloy (RHEA) coating has been developed to overcome the surface damage defects in Ti-6Al-4V alloys under wear and erosion conditions. A crack-free insitu nitride (AlCrTiNbMo)Nx RHEA coating were successfully prepared by the gas assisted laser nitriding. According to XRD, SEM and TEM analysis, AlCrTiNbMo coating was composed of single BCC phase, while (AlCrTiNbMo)Nx coating was constituted by BCC and TiN phases. The average microhardness of AlCrTiNbMo and (AlCrTiNbMo)Nx RHEA coatings were 591.2 HV0.3 and 1070.1 HV0.3 respectively. (AlCrTiNbMo)Nx RHEA coating has lower friction coefficient and wear rate than AlCrTiNbMo RHEA coating and the substrate from room temperature to 600 degrees C. At 600 degrees C, (AlCrTiNbMo)Nx RHEA coating showed a rare self-lubricating behavior, the friction coefficient decreased to 0.275 (61.9 % of the substrate), and the wear rate was 0.06 mm3/N mx 10-4 (1.24 % of the substrate). The erosion rates of (AlCrTiNbMo)Nx RHEA coating at erosion angle of 30 degrees, 60 degrees and 90 degrees were 11.0 %, 22.3 %, and 36.8 % of those of the substrate, respectively. The homogeneous and hard TiN array-BCC structure generated by in-situ self-reaction was the main factor for the excellent wear and erosion resistance of (AlCrTiNbMo)Nx RHEA coating. Thus, (AlCrTiNbMo)Nx RHEA coating and corresponding laser nitride technique were expected to be applied in the severe working environment of high temperature, wear, and erosion.

To solve the problem of data sparsity in recommendation system, this paper proposes an expert collaborative filtering algorithm integrating multimodal information. Firstly, the algorithm uses a multimodal feature extractor to capture the text features formed by expert review text and item text and the visual features extracted from item images. The item's multimodal information is effectively integrated using bilinear pooling method, and the multimodal representation of the item is obtained. Then, the multimodal representation is input into the neural network model together with user vectors and item vectors, and the user preference is represented by the Euclidean distance between user vectors and item vectors. At the same time, to avoid the phenomenon of "default positive comments" that may exist among users and the ambiguity of text sentiment, the paper extracts the sentiment score of text and user facial features and calculates their similarity to correct the rating matrix, making the rating matrix closer to the user's real preferences. The effectiveness of the model is verified on the MovieLens and Amazon Book multimodal datasets.

MgLi alloys have a wide application in the lightweight structure materials on account of their ultra-light-weight, good machinability and high specific strength. However, the low strength, insufficient wear and corrosion resistances of MgLi alloys have become a critical problem that hinders their applications. In order to achieve excellent wear and corrosion properties, LC-Cu3Al, LC-Cu6Al and LC-Cu9Al coatings are designed and prepared on MgLi alloy substrates in this study. The phase structure, microstructure, hardness, wear and corrosion properties of CuAl alloy coatings are investigated in detail. The results show that a good metallurgical bond is achieved between the CuAl alloy coatings and the substrates. LC-Cu6Al coating's hardness, wear and corrosion properties are the best. Compared with the MgLi alloy substrate, the average hardness of LC-Cu6Al coating increases by 6.75 times, while the wear volume decreases by 88.32 % and the corrosion current density decreases by two orders of magnitude. The results suggest that using laser cladding CuAl alloy coatings is an excellent strategy to enhance the surface properties of MgLi alloy.

The poor surface properties of Mg-Li alloys and the excellent comprehensive performance of high -entropy alloy (HEA) coatings are highly complementary. This study uses laser cladding technology to deposit an AlTiVNiCu/ Cu-Al gradient functional coating with favourable wear and corrosion properties on Mg-Li alloy substrates. Through the design of the gradient coating structure, laser cladding is used to coat HEA layers on low -meltingpoint Mg-Li alloys, which effectively reduces further dilution of Mg and Li in the substrate. Notably, the Mg content in the prepared AlTiVNiCu protective layer is only 2.34 %. Furthermore, the microstructure, phase structure, microhardness, tribological and corrosion properties of the gradient coating are systematically studied. The protective layer consists of V -rich BCC1, (Ni, Ti) -rich BCC2, Cu -rich FCC, Mg -rich BCC3 and nanoscale beta-Ti phases, with a predominance of body -centered cubic (BCC) structural phases. The average microhardness is 547.46 HV0.3. Under a load of 3 N, the average friction coefficient of the protective layer is 0.54, with a wear volume of only 2.87 % compared with that of the substrate, which is satisfactory wear resistance. In a 3.5 wt% NaCl solution, the protective layer exhibits the highest corrosion potential (-0.49 VSCE, where SEC denotes saturated calomel electrode), and the lowest corrosion current density (1.14 x 10-6 A & sdot;cm- 2), representing a remarkable enhancement compared with those of the substrate (-1.57 VSCE and 5.41 x 10-4 A & sdot;cm- 2).

AlCuTiVCr/Cu-Al lightweight high-entropy alloy gradient coating was successfully prepared by laser cladding to improve the corrosion resistance of Mg-Li alloy. The boundaries between AlCuTiVCr protective layer, Cu-Al transition layer and Mg-Li alloy substrate are distinct, forming an excellent metallurgical bonding. The gradient coating design has effectively balanced the thermal matching difference between HEA coating. The prepared AlCuTiVCr protective layer comprises (V, Cr)-rich BCC1, Mg-rich BCC2, AlCu2Ti and & beta;-Ti phases. The Cu-Al transition layer is composed of & beta;-Li, Cu2Mg and Al2Cu3 phases. The corrosion potential of AlCuTiVCr protective layer (-0.46 VSEC) is 1.09 VSEC higher than that of Mg-Li alloy (-1.55 VSEC), while the corrosion current density of the protective layer (9.41 x 10- 7 A cm-2) is about three orders of magnitude lower than that of the substrate (5.97 x 10-4 A cm-2).

Network embedding is a technique used to learn a low-dimensional vector representation for each node in a network. This method has been proven effective in network mining tasks, especially in the area of recommendation systems. The real-world scenarios often contain rich attribute information that can be leveraged to enhance the performance of representation learning methods. Therefore, this article proposes an attribute network embedding recommendation method based on self-attention mechanism (AESR) that caters to the recommendation needs of users with little or no explicit feedback data. The proposed AESR method first models the attribute combination representation of items and then uses a self-attention mechanism to compactly embed the combination representation. By representing users as different anchor vectors, the method can efficiently learn their preferences and reconstruct them with few learning samples. This achieves accurate and fast recommendations and avoids data sparsity problems. Experimental results show that AESR can provide personalized recommendations even for users with little explicit feedback information. Moreover, the attribute extraction of documents can effectively improve recommendation accuracy on different datasets. Overall, the proposed AESR method provides a promising approach to recommendation systems that can leverage attribute information for better performance.

This paper proposes an advanced expert collaborative filtering recommendation algorithm. Although ordinary expert system filtering algorithms have improved the recommendation accuracy of collaborative filtering technology to a certain extent, they have not screened the level of expertise of experts, and the credibility of experts varies. Therefore, this paper proposes an expert mining system based on signal fluctuations. The algorithm uses signal processing technology to filter the level of experts. This method introduces a kurtosis factor. Regarding the user's rating sequence as a random discrete signal, and then randomly sorting the user's ratings k times, the average kurtosis of the user is obtained. And take the kurtosis value as the credibility of expert users. Through experiments on multiple datasets including MovieLens, Jester, Booking-Crossings, and Last.fm, we have proved the advancement and reliability of our method.

The improvement of corrosion resistance and wear resistance levels is of great significance because of the farranging impacts of them on the further application of magnesium alloys. In this work, a novel TiB/Ti50Zr25Al15Cu10 medium-entropy alloy composite coating (MEACC) was successfully prepared on the surface of magnesium alloy by laser cladding. The effect of LaB6 content on microstructure, micro-hardness, wear resistance and corrosion resistance of Ti50Zr25Al15Cu10 coatings was studied in detail. The results show that TiB/Ti50Zr25Al15Cu10 MEACCs are mainly composed of alpha-Ti, BCC and TiB phases. With LaB6 content increased, the microstructure of the coating is refined and homogenized. And when the addition amount is excessive, Cu2La clusters appear. The lattice distortion caused by La3+ with large ion radius leading to the formation of hard BCC solid solution, as well as the dispersion strengthening of TiB in-situ, the coating presents high micro-hardness and excellent wear resistance. Especially, the average micro-hardness values of Ti50Zr25Al15Cu10 coating with 4 wt% LaB6 is similar to 679.4 HV0.3, which is about 11 times than that of the substrate (similar to 61.6 HV0.3). Besides, TiB/Ti50Zr25Al15Cu10 MEACC displays excellent corrosion resistance. The minimum corrosion current density is 8.52 x 10(-7) A center dot cm(-2), and the oxidation phenomenon emerged in the coating. The oxide film is mainly composed of the oxidation state of Ti, Zr, Al, Cu and the sub-oxide state of Cu.

Aspect-based sentiment analysis (ABSA) aims to predict the sentiment polarity with the different aspect terms or categories, which play an important role to guide the representation of context vector. Previous studies have used concatenation operation as a common means of information aggregation, which increase irrelevant noise and lose the dependence between the original features. In this paper, we propose a lightweight feature enhanced dual-GRU to selectively learn the feature relevance between aspect terms and context. The dual-GRU contains an extended aspect-related GRU and a position-related GRU to generate relevant information adaptively. Meanwhile, we construct a context-related GRU to enhance the dependency between aspect terms and context. Extensive experimental results demonstrate that the proposed model is reliable and effective in improving the performance of the two tasks of ABSA.

With the development of network technology, the amount of information on the network has grown and expanded rapidly with an exponential law, and its information organization is heterogeneous, diverse, and distributed. With the vigorous development of Internet information services, the scale of its information resources has also exploded. For ordinary users, the problems of “information trek” and “information overload” on the Internet are becoming increasingly serious. In order to solve the problem of information overload, recommendation system has become an indispensable tool for today’s e-commerce platform, which can help users find valuable information quickly. Collaborative filtering-based recommendation algorithms have been widely applied and studied in recommendation systems. Although the collaborative filtering algorithm has been widely used, there are still problems such as data sparsity, scalability, and cold start, which seriously limit the quality of recommendations. Therefore, collaborative filtering algorithms face many challenges. Especially recommendation systems using collaborative filtering technology. This article discusses the cold start system, the cold start user, and the cold start scenario. Effective use of project content information and user personal information is one of the effective methods to solve the cold start problem, that is, a hybrid recommendation technology combining information filtering and collaborative filtering. This paper proposes a hybrid recommendation technology that can better solve the cold start problem.

The basis of the identification of network security situation element is to perform the feature extraction of situation data effectively. Considering the problem that the Back Propagation(BP) neural networks have excessive dependence on data labels when it has a learning of massive security situation information data, a network security situation element identification method is proposed, which combines deep stack encoder and BP algorithm. It trains the network layer by layer through unsupervised learning algorithm. On this basis the deep track encoder by stacking can be obtained. The unsupervised training of the network is realized when using the encoder to extract the characteristic of the data sets. It is verified by simulation experiments that the method can improve the performance and accuracy of situational awareness effectively.

The role identification in social networks is important for analyzing and understanding social networks, predicting user behavior, and researching relationships and interactions between users. Most role identification works are based on link analysis following the ideas of PageRank and HITS, and some methods also combing link analysis and content analysis. However, previous role identification methods focus on the amount of topic released by the target user without considering the importance of the topic. In order to identify high-value users by topic weights in social networks, this paper first proposes a method of assigning topic weights based on the LDA model and the collective credit allocation method in science(CCA). Then we introduce content value density based on the Kullback-Leibler divergence of topic weight and topic distribution to rank the high-value users. In addition, the new method is more suitable for calculation on large-scale networks and time-evolution networks because the new method is an inductive method which works on the local graph. Finally, the experiments are carried out to analyze the effect of the topic weight allocation and content value density in a real data set. The results show that the new method is superior to the compared methods when identifying high-value users in certain circumstances.