Search

To address the problems in the existing video super-resolution methods, such as noise, over smooth and visual artifacts, which are caused by reliance on limited external training or mismatch of internal similarity instances, this study proposes a video super-resolution reconstruction algorithm based on deep learning and spatio-temporal feature similarity (DLSS-VSR). The video super-resolution reconstruction mechanism with joint internal and external constraints is established utilizing both external deep correlation mapping learning and internal spatio-temporal nonlocal self-similarity prior constraint. A deep learning model based on deep convolutional neural network is constructed to learn the nonlinear correlation mapping between low-resolution and high-resolution video frame patches. A spatio-temporal feature similarity calculation method is proposed, which considers both internal video spatio-temporal self-similarity and external clean nonlocal similarity. For the internal spatio-temporal feature self-similarity, we improve the accuracy and robustness of similarity matching by proposing a similarity measure strategy based on spatio-temporal moment feature similarity and structural similarity. The external nonlocal similarity prior constraint is learned by patch group-based Gaussian mixture model. The time efficiency for spatio-temporal similarity matching is further improved based on saliency detection and region correlation judgment strategy. Experimental results demonstrate that the DLSS-VSR achieves competitive super-resolution quality compared to other state-of-the-art algorithms.

Previous studies have identified two emission modes in PSR B1859+07: a normal mode that has three prominent components in the average profile, with the trailing one being the brightest, and an anomalous mode (i.e. the A mode) where emissions seem to be shifted to an earlier phase. Within the normal mode, further analysis has revealed the presence of two sub-modes, i.e. the cW mode and cB mode, where the central component can appear either weak or bright. As for the anomalous mode, a new bright component emerges in the advanced phase while the bright trailing component in the normal mode disappears. New observations of PSR B1859+07 by using the Five-hundred-meter Aperture Spherical radio Telescope (FAST) have revealed the existence of a previously unknown emission mode, dubbed as the Af mode. In this mode, all emission components seen in the normal and anomalous modes are detected. Notably, the mean polarization profiles of both the A and Af modes exhibit an orthogonal polarization angle jump in the bright leading component. The polarization angles for the central component in the original normal mode follow two distinct orthogonal polarization modes in the A and Af modes respectively. The polarization angles for the trailing component show almost the same but a small systematic shift in the A and Af modes, roughly following the values for the cW and cB modes. Those polarization features of this newly detected emission mode imply that the anomalous mode A of PSR B1859+07 is not a result of ``phase shift" or ``swooshes" of normal components, but simply a result of the varying intensities of different profile components. Additionally, subpulse drifting has been detected in the leading component of the Af mode.

We investigate the single-pulse emission variations of two pulsars, PSRs J0211+4235 and J0553+4111, observed with the Five-hundred-meter Aperture Spherical Radio Telescope at the 1.25 GHz central frequency. The observation sessions span from December 2020 to July 2021, with 21 and 22 observations for them. The integrated pulse profile of PSR J0211+4235 shows that there is a weak pulse component following the main component. And for PSR J0553+4111, it shows a bimodal profile with a bridge component in the middle. PSR J0211+4235 presents significant nulling phenomenon with nulling dura- tion lasting from 2 to 115 pulses and burst duration lasting from 2 to 113 pulses. The nulling fraction of each observation is determined to be 45% to 55%. No emission greater than three sigma is found in the mean integrated profile of all nulling pulses. In most cases, the pulse energy changes abruptly during the transition from null to burst, while in the transition from burst to null there are two trends: abrupt and gradual. We find that the nulling phenomenon of PSR J0211+4235 is periodic by Fourier transform of the null and burst state. In addition, the single-pulse modulation characteristics of these two pulsars are investigated, and the dis- tribution of modulation index, LRFS, 2DFS are analyzed with PSRSALSA. The left peak of PSR J0553+4111 has intensity modulation. Finally, the polarization property of these two pulsars are obtained through polarization calibration, and their characteristics are analyzed. The possible physical mechanisms of these phenomena are discussed.

Scientific literature retrieval provides convenience for researchers to find scientific literature related to the query. It is an important part of scientific research to search related papers given a paper title as query. However, for scientific literature retrieval tasks, most of the existing retrieval methods do not consider sentence-level semantic matching so that the retrieval performance is limited. With the success of neural networks, neural information retrieval methods have been widely studied and achieved good retrieval results. In this paper, we propose a semi-supervised semantic-enhanced scientific literature retrieval framework. The framework is composed of two networks: a self-attention convolutional encoder-decoder network and a sentence-level attention scientific literature retrieval network. By joint training of the two networks, the proposed semi-supervised semantic-enhanced scientific literature retrieval model can fully capture the rich semantic information of scientific text data and leverages human labeled scientific text data to improve the discriminativeness of the learned semantic representation. The retrieval results on two scientific literature datasets demonstrate that the proposed method significantly and consistently outperforms the other baseline methods.

Cross-media search from large-scale social network big data has become increasingly valuable in our daily life because it can support querying different data modalities. Deep hash networks have shown high potential in achieving efficient and effective cross-media search performance. However, due to the fact that social network data often exhibit text sparsity, diversity, and noise characteristics, the search performance of existing methods often degrades when dealing with this data. In order to address this problem, this article proposes a novel end-to-end cross-media semantic correlation learning model based on a deep hash network and semantic expansion for social network cross-media search (DHNS). The approach combines deep network feature learning and hash-code quantization learning for multimodal data into a unified optimization architecture, which successfully preserves both intramedia similarity and intermedia correlation, by minimizing both cross-media correlation loss and binary hash quantization loss. In addition, our approach realizes semantic relationship expansion by constructing the image-word relation graph and mining the potential semantic relationship between images and words, and obtaining the semantic embedding based on both internal graph deep walk and an external knowledge base. Experimental results demonstrate that DHNS yields better cross-media search performance on standard benchmarks.

Bursty topic discovery aims to automatically identify bursty events and continuously keep track of known events. The existing methods focus on the topic model. However, the sparsity of short text brings the challenge to the traditional topic models because the words are too few to learn from the original corpus. To tackle this problem, we propose a Sparse Topic Model (STM) for bursty topic discovery. First, we distinguish the modeling between the bursty topic and the common topic to detect the change of the words in time and discover the bursty words. Second, we introduce “Spike and Slab” prior to decouple the sparsity and smoothness of a distribution. The bursty words are leveraged to achieve automatic discovery of the bursty topics. Finally, to evaluate the effectiveness of our proposed algorithm, we collect Sina weibo dataset to conduct various experiments. Both qualitative and quantitative evaluations demonstrate that the proposed STM algorithm outperforms favorably against several state-of-the-art methods

With the significant development of the Internet of Vehicles (IoV), various modal data, such as image and text, are emerging, which provide data support for good vehicle networking services. In order to make full use of the cross-modal data, we need to establish a common semantic representation to achieve effective measurement and comparison of different modal data. However, due to the heterogeneous distributions of cross-modal data, there exists a semantic gap between them. Although some deep neural network (DNN) based methods have been proposed to deal with this problem, there still exist several challenges: the qualities of the modality-specific features, the structure of the DNN, and the components of the loss function. In this paper, for representing cross-modal data in IoV, we propose a common semantic representation method based on object attention and adversarial learning (OAAL). To acquire high-quality modality-specific feature, in OAAL, we design an object attention mechanism, which links the cross-modal features effectively. To further alleviate the heterogeneous semantic gap, we construct a cross-modal generative adversarial network, which contains two parts: a generative model and a discriminative model. Besides, we also design a comprehensive loss function for the generative model to produce high-quality features. With a minimax game between the two models, we can construct a shared semantic space and generate the unified representations for cross-modal data. Finally, we apply our OAAL on retrieval task, and the results of the experiments have verified its effectiveness.

Learning to rank has been applied to many web searches, but it cannot be directly applied to retrieval scientific research projects. In scientific research project query, people are not only concerned about the name of the project, but also the digital information and side-information, such as duration, achievements, funding amount of this project. Howerver, the existing learning to rank methods ignore the side-information of scientific research projects. Therefore, we propose a Side-information Pointwise Ranking model (SIPR) for scientific research project query based on deep language model, click model and learning to rank. First, we use the deep language model to extract the semantic information of text, and design a relevance calculation model to extract features of side-information, then we merge the above two features. After that we use the click model to eliminate position bias, and get a ranking score through pointwise DNN. Finally, we can get the query results ordered by these scores. Experiments on real scientific research projects dataset demonstrate that our model can achieve better performance.

Rotating Radio Transients are a relatively new subclass of pulsar characterized by sporadic bursting emission of single pulses. Here, we present a single-pulse analysis of a rotating radio transient, RRAT J0139+3336, using Five-hundred-meter Aperture Spherical radio Telescope at 1250 MHz. Within 3.32 hr of continuous observation, 152 single pulses were detected in RRAT J0139+3336, with the pulse rate of 45 pulses per hour. We perform a spectral analysis on the single pulses of this pulsar for the first time, finding its mean spectral indices to be −3.2 ± 0.2, which is steeper than most known pulsars. On a single-pulse basis, we produce the first polarimetric profile of this pulsar, which fits well with the rotating vector model. The single pulses are clearly affected by diffractive scintillation with a characteristic scintillation bandwidth of v sc = 28 ± 9 MHz. The pulse energy distribution for RRAT J0139+3336 can be described by a log-normal model.

Short text analysis is a challenging task as far as the sparsity and limitation of semantics. The semantic extension approach learns the meaning of a short text by introducing external knowledge. However, for the randomness of short text descriptions in microblogs, traditional extension methods cannot accurately mine the semantics suitable for the microblog theme. Therefore, we use the prominent and refined hashtag information in microblogs as well as complex social relationships to provide implicit guidance for semantic extension of short text. Specifically, we design a deep hash model based on social and conceptual semantic extension, which consists of dual semantic extension and deep hashing representation. In the extension method, the short text is first conceptualized to achieve the construction of hashtag graph under conceptual space. Then, the associated hashtags are generated by correlation calculation based on the integration of social relationships and concepts to extend the short text. In the deep hash model, we use the semantic hashing model to encode the abundant semantic features and form a compact and meaningful binary encoding. Finally, extensive experiments demonstrate that our method can learn and represent the short texts well by using more meaningful semantic signal. It can effectively enhance and guide the semantic analysis and understanding of short text in microblogs.

With the rapid development of social network platforms, more and more people are using them to search for material related to their interests. As the texts of social media messages are usually so short, when traditional existing document modeling methods are used in social network search tasks, the problem of semantic sparsity arises, leading to low-quality semantic representation and low-precision social network search results. Fortunately, besides of short text, social media data also has other features, such as timestamps, locations, and its user information. In light of this, to realize precise social network search, we propose a multi-feature probabilistic graphical model (MFPGM), which can generate high-quality semantic representation. To deal with the problem of semantic sparsity, we exploit two strategies in MFPGM. First, we propose a concept named special region and utilize location information to aggregate short text into long text. Second, we introduce the biterm pattern that can generate dense semantic space by supposing that a biterm occurring in the same context has the same topic. In order to generate high-quality semantic representations, we simultaneously model multiple features (i.e., biterm, user, location and timestamp) of social network data to enhance the semantic learning process of MFPGM. We conduct a lot of experiments on real-word datasets, and the comparisons with several state-of-art baseline methods have demonstrated the superiority of our MFPGM on topic quality and search performance. Additionally, with the help of the generated semantic representations, MFPGM allows people to analyze the relationships between time and the popularities of topics.

Computer science discipline includes many research fields, which mutually influence and promote each other’s development. This poses two great challenges of predicting the research topics of each research field. One is how to model fine-grained topic representation of a research field. The other is how to model research topic of different fields and keep the semantic consistency of research topics when learning the scientific influence context from other related fields. Unfortunately, the existing research topic prediction approaches cannot handle these two challenges. To solve these problems, we employ multiple different Recurrent Neural Network chains which model research topics of different fields and propose a research topic prediction model based on spatial attention and semantic consistency-based scientific influence modeling. Spatial attention is employed in field topic representation which can selectively extract the attributes from the field topics to distinguish the importance of field topic attributes. Semantic consistency-based scientific influence modeling maps research topics of different fields to a unified semantic space to obtain the scientific influence context of other related fields. Extensive experiment results on five related research fields in the computer science (CS) discipline show that the proposed model is superior to the most advanced methods and achieves good topic prediction performance.

Internet of Things search has great potential applications with the rapid development of Internet of Things technology. Combining Internet of Things technology and academic search to build academic search framework based on Internet of Things is an effective solution to realize massive academic resource search. Recently, the academic big data has been characterized by a large number of types and spanning many fields. The traditional web search technology is no longer suitable for the search environment of academic big data. Thus, this paper designs academic search framework based on Internet of Things Technology. In order to alleviate the pressure of the cloud server processing massive academic big data, the edge server is introduced to clean and remove the redundancy of the data to form a clean data for further analysis and processing by the cloud server. Edge computing network effectively makes up for the deficiency of cloud computing in the conditions of distributed and high concurrent access, reduces long-distance data transmission, and improves the quality of network user experience. For Academic Search, this paper proposes a novel weakly supervised academic search model based on knowledge-enhanced feature representation. The proposed model can relieve high cost of acquisition of manually labeled data by obtaining a lot of pseudolabeled data and consider word-level interactive matching and sentence-level semantic matching for more accurate matching in the process of academic search. The experimental result on academic datasets demonstrate that the performance of the proposed model is much better than that of the existing methods.

We present simultaneous broad-band radio observations on the abnormal emission mode from PSR B1859+07 using the Five-hundred-meter Aperture Spherical radio Telescope (FAST). This pulsar shows peculiar emission, which takes the form of occasional shifts of emission to an early rotational phase and mode change of emission at the normal phase. We confirm all these three emission modes with our data sets, including the B (burst) and Q (quiet) modes of the non-shifted pulses and the emission shift mode with a quasi-periodicity of 155 pulses. We also identify a new type of emission shift event, which has emission at the normal phase during the event. We studied polarization properties of these emission modes in detail, and found that they all have similar polarization angle curve, indicating the emissions of all these three modes are from the same emission height.

Since the research of sentiment analysis is mostly concentrated in the field of sentiment analysis on Weibo, and there is less research on sentiment analysis of financial text, this thesis proposes a financial sentiment analysis model based on pre-training and TextCNN. First, the pre-trained model is used to initially extract the emotional features of the text. It can extract text features well, and can extract information between words at arbitrary intervals when processing text sequences. Then use the improved TextCNN to construct a sentiment analysis network to further extract the sentiment features of the text, effectively identify the sentiment of the text, and complete the sentiment analysis of financial text. This thesis conducts experiments on a balanced corpus data set based on financial texts, and compares it with other classic sentiment analysis algorithms. Experimental results show that the proposed method works best in the field of financial text sentiment analysis.

In order to help investors get effective information from a large amount of financial news data as soon as possible, a topic detection model based on multi-view text semantics and clustering topic detection algorithm is proposed. In the financial news data set, different models are used to extract the characteristics of the news, and the characteristics of various models are merged. The clustering algorithm is improved by introducing JS divergence and time decay factors. The experimental results show that compared with the traditional topic detection model, the proposed method has higher accuracy of topic detection and shorter runtime of clustering algorithm.

With the development of the economy, more and more people on social platforms share financial-related information and hope to understand the development of financial markets through relevant financial data. Aiming at this problem, this thesis proposes a sentiment analysis algorithm of financial texts based on semantic matching. First, preliminary text vectorization is performed on financial text data. Then further training and fine-tuning are performed through pre-trained models in order to fully tap the associations between text contexts and better grasp the semantic focus, so as to better model the text content and obtain higher-level financial text semantic representation. Next, through the improved Siamese network semantic matching model, financial text vectors are trained for semantic matching, so that the distance between financial text vectors with the same emotional category is closer and the distance between different classes is farther, which further optimizes the semantic representation of financial text. Finally, support vector machines are used as classifier to perform sentiment classification of financial texts. Experimental results show that compared with other classic sentiment analysis algorithms, the proposed algorithm has the best sentiment analysis effect on financial text.

Stock trend prediction has always been the focus of research in the field of financial big data. Stock data is complex nonlinear data, while stock price is changing over time. Based on the characteristics of stock data, this paper proposes a financial big data stock trend prediction algorithm based on attention mechanism (STPA). We adopt Bidirectional Gated Recurrent Unit (BGRU) and attention mechanism to capture the long-term dependence of data on time series. The attention mechanism is used to analyze the weight of the impact of data from different time periods on the trend prediction results, thereby reducing the error of stock data change trend prediction and improving the accuracy of trend prediction. We select the daily closing price data of 10 stocks for model training and performance evaluation. Experimental results demonstrate that the proposed method STPA achieves higher precision, recall rate and F1-Score in predicting stock change trends than the other methods. Compared with mainstream methods, STPA improves the precision by 4%, improves recall by 2.5%, and improves F1-Score by 3.2%.

Social networks such as Twitter, Facebook, and Sina microblogs have emerged as major sources for discovering and sharing the latest topics. Because social network topics change quickly, developing an effective method to model such topics is urgently needed. However, topic modeling is challenging due to the sparsity problem and the dynamic change of topics in microblog streams. In this study, we propose dynamic topic modeling via a self-aggregation method (SADTM) that can capture the time-varying aspect of topic distributions and resolve the sparsity problem. The SADTM aggregates the observable and unordered short texts as the aggregated document without leveraging an external context to overcome the sparsity problem of short text. Furthermore, we exploit word pairs instead of words for each microblog to generate more word co-occurrence patterns. The SADTM models temporal dynamics by using the topic distribution at previous time steps in microblog streams to infer the current topic from sequential data. Extensive experiments on a real-world Sina microblog dataset demonstrate that our SADTM algorithm outperforms several state-of-the-art methods in topic coherence and cluster quality. Additionally, when applied in a search scene, our SADTM significantly outperforms all baseline methods in terms of the quality of the search results.

Given the social network short text native sparsity, semantic inference becomes an infeasible task for conventional topic models. By exploiting the spatial and temporal characteristics of social network data, we propose a social network short text semantic modeling method, named by Spatial and Temporal Topic Model (STTM). To further overcome short text sparsity, STTM leverages co-occurrence word–word pair to reduce the sparsity problem, and moreover, it incorporates time information into the process of topics modeling in order to generate topics with higher quality. Experimental results over four real social media datasets verify the effectiveness of STTM.

Searching for microblog short text by their meaning is a challenging task because of the semantic sparsity of the information in social networks. The extended search approaches are commonly accepted which facilitate short text understanding and search by enriching the short text. However, they only analyze the literal semantics of short text, and the unique social characteristics of social network which also contain semantic information are not utilized well. To better capture the rich semantics in microblog short text, we propose a new microblog short text extended search method based on a semantic hashtag graph by combining social and conceptual information, which enriches each short text by concepts and associated hashtags to represent whole semantic features. Considering the microblog context, we introduce concepts through Wikipedia, as well as semantic consistency of hashtags. Specifically, for conceptual semantics, we propose a conceptual analysis method which merges explicit and implicit information in Wikipedia. For social semantics in hashtags, a semantic hashtag graph which combines social and conceptual information is put forward to generate semantic associated hashtags. We conduct experiments and the results show that our method is obviously better than the other existing state-of-the-art approaches in semantic understanding and search of short text.

Hashtag recommendation for microblogs is a very hot research topic that is useful to many applications involving microblogs. However, since short text in microblogs and low utilization rate of hashtags will lead to the data sparsity problem, it is difficult for typical hashtag recommendation methods to achieve accurate recommendation. In light of this, we propose HRMF, a hashtag recommendation method based on multi-features of microblogs in this article. First, our HRMF expands short text into long text, and then it simultaneously models multi-features (i.e., user, hashtag, text) of microblogs by designing a new topic model. To further alleviate the data sparsity problem, HRMF exploits hashtags of both similar users and similar microblogs as the candidate hashtags. In particular, to find similar users, HRMF combines the designed topic model with typical user-based collaborative filtering method. Finally, we realize hashtag recommendation by calculating the recommended score of each hashtag based on the generated topical representations of multi-features. Experimental results on a real-world dataset crawled from Sina Weibo demonstrate the effectiveness of our HRMF for hashtag recommendation.

The opinion leader mining is critical to the orderly and healthy development of social networks. The traditional method mostly mines opinion leaders from the perspective of users, and ignores the relationship between Weibo content, so the performance of opinion leader mining algorithm is limited. To solve this problem, this paper proposes an opinion leader mining algorithm based on the designed double-graph model, namely DGRank. Based on the improvement of the LeaderRank algorithm, we combine the relationship between users in social network, the number of fans, the number of Weibo and the number of followers to build a user graph model. Then, we calculate the similarity between the Weibo published by the users and build a Weibo graph model. The correlation between Weibo supplements the relationship between users, and the two models are merged into a unified double-graph model. Finally, we calculate the user impact score to find the opinion leader. The experimental results show that the proposed DGRank algorithm is superior to the traditional methods, and can more effectively dig out opinion leaders.

With the development of social networks, more and more people use social networks to publish and disseminate national security emergencies. In order to effectively control the spread and development of Chinese social network national security emergencies, we need to make an effective portrait of the emergencies. However, Chinese social network information has two research difficulties, such as text irregularity and few data sets in related fields, which may result in inaccurate event portrait results. In order to solve the above problems, we propose an algorithm based on the attention mechanism of Chinese part-of-speech tagging results (BLTAC) to perform emergency event portrait of Chinese social networks, which can efficiently perform emergency portraits. The BLTAC algorithm can be used to extract the Chinese social network emergency text entity name, and use the extracted entity name to describe the emergency event to perform event portrait. The experimental results show that the F1-score of our algorithm for the entity names recognition in each category on the Weibo dataset is improved compared with the other methods.

Bursty topic spreads very quickly and generate huge influence in Weibo. Therefore, bursty topic detection is one of the hot spots of topic detection and tracking. Most of the existing bursty topic detection methods do not consider the basic weight of the bursty term and the filtration of the invalid bursty term. In this paper, we propose a bursty topic detection method BTDF based on calculation of bursty term value and recognition of pseudo bursty term. The proposed BTDF uses topic models and clustering methods to get general topics, and identifies sudden topics from general topics by judging whether topic keywords contain bursty terms. In BTDF, we extract the bursty term by using the basic weight and bursty weight of the term and filter the pseudo bursty terms by analyzing the novelty of the terms. The experiments conducted on Weibo data show that the proposed method achieves better performance in bursty topic detection.

Short text sparsity, oral language, and polysemy are the main problems when dealing with social network data, which make the traditional methods hard to obtain the true meaning of social network data. Due to the above issues, topic detection for social network data is not that easy. And to solve the above problems, we propose an original Clustering Algorithm based on Semantics, Time, and Social relationship (CASTS) for topic detection. Firstly, to overcome short text sparsity and polysemy problems, the CASTS leverages the Bidirectional Encoder Representations from Transformers (BERT), which can pre-train on large-scale social network short text data to obtain concise text representation with rich semantics. Secondly, by combining the short text representation, time, and social relationship, the CASTS can efficiently detect topics. Finally, we conduct experiments on Weibo dataset to verify the correctness and effectiveness of CASTS.

Because of everyone's involvement in social networks, social networks are full of massive multimedia data, and events are got released and disseminated through social networks in the form of multi-modal and multi-attribute heterogeneous data. There have been numerous researches on social network search. Considering the spatio-temporal feature of messages and social relationships among users, we summarized an overall social network search framework from the perspective of semantics based on existing researches. For social network search, the acquisition and representation of spatio-temporal data is the basis, the semantic analysis and modeling of social network cross-media big data is an important component, deep semantic learning of social networks is the key research field, and the indexing and ranking mechanism is the indispensable part. This paper reviews the current studies in these fields, and then main challenges of social network search are given. Finally, we give an outlook to the prospect and further work of social network search.

Image is an important part of microblog, and its visual information can offer additional semantics besides the textual information. To overcome short text’s semantic sparsity problem and fully utilize the semantics of text and image, we propose a cross-modal short text expansion method for microblog search in this paper. First, we expand short texts using the distributed representations of words, and then based on deep neural network, we extract related information of images and append them to the original short text. The expanded pseudo-documents contain richer semantics, and by turning pseudo-documents into vectors, we can achieve accurate microblog search. Experiments on real-world datasets show that the proposed cross-modal short text expansion method can effectively extract the semantics of microblogs and improve search performance.

Social network generates massive text data every day, which makes it important to mine its semantics. However, due to the inability to combine global semantics with local semantics, existing semantic modeling methods cannot overcome the sparseness of short texts and the ambiguity of words in different spatial-temporal environments. In this paper, we propose a semantic modeling method for social network short text, named Spatial-temporal topic embedding (STTE), which combines the spatial-temporal global context information and local context information. We first design a topic model that utilizes the text feature, time feature and location feature at the same time to generate accurate spatial-temporal global context information. Then, we employ this global information to predict an explicit topic for each word and regard the combination of each word and its assigned topic as a new pseudo word. After that, we exploit pseudo word sequence as the input of embedding vector model and finally learn the text feature which could reflect the text semantic with social network characteristics. Classification and search experiments in real-world datasets of the social network have verified that the proposed STTE has better semantic modeling ability than other baseline methods.