Social text geographic position reasoning method and device based on multi-feature fusion and electronic equipment

Abstract

The embodiment of the invention discloses a social text geographic position reasoning method and device based on multi-feature fusion and electronic equipment. A specific embodiment of the method comprises the following steps: acquiring latitude and longitude data of a social text site; performing geographic grid division on the longitude and latitude data to generate a geographic number grid; constructing a semantic feature vector according to pre-acquired social text semantic information; constructing a geographic feature vector according to pre-acquired urban population data; and inputting the semantic feature vector and the geographic feature vector into a geographic position reasoning model to obtain a geographic position reasoning result, the geographic position reasoning result comprising a grid number. According to the embodiment, the reasoning precision on different scales can be improved, and the social text geographic position reasoning problem is solved.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the fields of computer technology, social media text processing, geospatial entity recognition, and deep learning technology, and specifically to a social text geolocation inference method, device, and electronic device based on multi-feature fusion. Background Art

[0002] With the rapid development of internet technology, social media platforms have become an essential part of modern life. These platforms not only attract vast user communities but also generate a vast amount of user-generated content (text, images, videos, etc.), providing a rich data resource for geographic information mining and location-based services. While this data contains rich geospatial information, only 1% of it is geo-located, significantly limiting its effective utilization.

[0003] Traditional geolocation mining methods primarily rely on IP (Internet Protocol) address positioning and GPS (Global Positioning System) data. However, these methods often face legal and ethical challenges in practical application regarding user privacy protection. Without active user cooperation, their effectiveness is often significantly reduced. In recent years, with the rapid development of big data and artificial intelligence technologies, geolocation inference techniques based on machine learning and deep learning have achieved significant breakthroughs. Existing research focuses on two main approaches: document-level geocoding, which infers the potential geolocation of user posts by analyzing textual linguistic features; and geographic entity-based georesolution, which achieves accurate location inference through a process of place name recognition, disambiguation, and encoding. These two types of research can be divided into those that leverage implicit and explicit geographic information in text to map unstructured text to geographic coordinates.

[0004] Researchers typically prioritize and utilize explicit geographic information in text, such as specific place names, landmarks, or address descriptions, to achieve geolocation inference in text through place name disambiguation and geocoding. However, implicit geographic information, such as contextual cues, cultural background knowledge, or specific linguistic conventions (such as dialects and slang), is often not directly accessible from the text itself and requires in-depth analysis using more sophisticated natural language processing techniques. For explicit geographic information such as mined place names, this approach uses contextual clues as rules to disambiguate place names and then obtain geocoding, enabling geolocation inference in text. The core principle of this approach is to construct place name disambiguation rules using multidimensional features: integrating statistical features such as population density and word type, while also establishing association constraints based on spatial topological relationships between locations. With the rapid development of representation learning technologies, location inference methods based on word embeddings and neural networks have emerged. Researchers use language models to mine implicit geographic features in text, enabling inference of users' locations. In the specific implementation, researchers constructed multidimensional feature vectors by extracting event locations, keywords and time series features, and performed feature clustering analysis based on spatial distribution, time series and topic correlation to achieve geographic location inference.

[0005] However, these approaches still face the following challenges: First, they struggle to accurately identify geographic entities and understand text semantics when processing short social texts; second, their inference accuracy remains low across different scales. The above analysis demonstrates that addressing the sparsity of geographic information in short social texts is crucial, and on this basis, improving the accuracy of location inference at different spatial scales.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0007] The content of this disclosure is used to briefly introduce concepts that will be described in detail in the detailed description section below. The content of this disclosure is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0008] Some embodiments of the present disclosure propose a social text geographic location inference method, device, and electronic device based on multi-feature fusion to solve the technical problems mentioned in the above background technology section.

[0009] In a first aspect, some embodiments of the present disclosure provide a method for social text geographic location inference based on multi-feature fusion, the method comprising: obtaining latitude and longitude data of the social text location; performing geographic grid division on the above latitude and longitude data to generate a geographic number grid, wherein each grid in the geographic number grid corresponds to a grid number; constructing a semantic feature vector based on pre-acquired social text semantic information; constructing a geographic feature vector based on pre-acquired city population data; inputting the above semantic feature vector and the above geographic feature vector into a geographic location inference model to obtain a geographic location inference result, wherein the geographic location inference result includes a grid number.

[0010] In a second aspect, some embodiments of the present disclosure provide a social text geographic location inference device based on multi-feature fusion, the device comprising: an acquisition unit configured to acquire the latitude and longitude data of the social text location; a grid division unit configured to perform geographic grid division on the above-mentioned latitude and longitude data to generate a geographic numbered grid, wherein each grid in the geographic numbered grid corresponds to a grid number; a first construction unit configured to construct a semantic feature vector based on pre-acquired social text semantic information; a second construction unit configured to construct a geographic feature vector based on pre-acquired urban population data; an input unit configured to input the above-mentioned semantic feature vector and the above-mentioned geographic feature vector into a geographic location inference model to obtain a geographic location inference result, wherein the geographic location inference result includes a grid number.

[0011] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0012] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation of the first aspect is implemented.

[0013] The above-described embodiments of the present disclosure have the following beneficial effects: The multi-feature fusion-based geolocation inference method for social text, as described in some embodiments of the present disclosure, can improve inference accuracy at different scales and address the geolocation inference problem in social text. Given that deep learning models can achieve a good understanding of social text and fully capture multiple aspects of text features, which is highly suitable for the geolocation inference task in social text, the present invention proposes a multi-feature fusion geolocation inference method for social text to address the geolocation inference problem in social text. This method utilizes a multi-layered feature extraction and integration mechanism, starting with the disambiguation of place-name entities and understanding text semantics, to infer the actual geolocation referenced by the text. The actual geographic space is then divided into a geographic network based on latitude and longitude intervals, ensuring both accuracy and computational efficiency. To achieve geolocation inference in social text, it is necessary to exploit as many features as possible in the existing data. The features mined through feature engineering primarily include semantic and geographic features. Finally, the geolocation inference model transforms the original fusion vector through a layer-by-layer feature extraction mechanism to learn high-level semantic representations and spatial distribution features, thereby achieving efficient and accurate geolocation inference. Thus, the reasoning accuracy at different scales is improved, and the problem of geolocation reasoning in social text is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0015] Figure 1 This is a framework diagram corresponding to the social text geographic location inference method based on multi-feature fusion according to the present disclosure; Figure 2 is a flowchart of some embodiments of the social text geographic location inference method based on multi-feature fusion according to the present disclosure; Figure 3 It is a schematic diagram of geographic grid division; Figure 4 It is a flowchart for enhancing corpus construction; Figure 5 It is a flowchart of feature extraction and position reasoning.

[0016] Figure 6 is a schematic structural diagram of some embodiments of a social text geographic location inference device based on multi-feature fusion according to the present disclosure; Figure 7 It is a structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0017] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0018] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other.

[0019] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0020] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0021] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0022] Before performing operations such as the collection, storage, and use of the information involved in this disclosure (such as the latitude and longitude data of social text locations, social text semantic information, etc.), relevant organizations or individuals must fulfill obligations such as conducting personal information security impact assessments, fulfilling the obligation to inform the personal information subjects, and obtaining the prior authorization and consent of the personal information subjects.

[0023] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0024] Figure 1 This is a framework diagram corresponding to the social text geographic location inference framework method based on multi-feature fusion according to the present disclosure.

[0025] exist Figure 1 In the framework diagram, first, the longitude and latitude data of the social text location can be obtained. Then, the longitude and latitude data can be divided into geographic grids to generate geographic number grids, where each grid in the geographic number grid corresponds to a grid number. Then, based on the pre-acquired social text semantic information (such as Figure 1Here, we can extract the features of the original text corpus and construct the semantic feature vector. For example, we can identify the place name (such as Xishan Linyu Community), identify the context information (some residents of Xishan Linyu Community), and enhance the corpus (Xishan Neighborhood Community North Gate). Then, we can use the pre-acquired urban population data (such as Figure 1 Finally, the geographic number grid, semantic feature vector, and geographic feature vector are input into the geographic location inference model (e.g., through three fully connected layers (Dense1, Dense2, and Dense3) for feature extraction) to obtain geographic location inference data. Here, the input dimension of the Dense1 layer can be t×h. The input dimension of the Dense1 layer can be h×s. The input dimension of the Dense1 layer can be s×v. The geographic location inference data includes the grid number (e.g., Figure 1 In the location inference diagram (for example, the output dimension is 124×101). In addition, social text location inference is a technical method that uses deep learning and other methods to infer the location of users' social texts when the social dynamics they post lack a clear location.

[0026] Next, see Figure 2 , Figure 2 A flowchart 200 is shown of some embodiments of a method for inferring a geographical location of a social text based on multi-feature fusion according to the present disclosure. The method for inferring a geographical location of a social text based on multi-feature fusion includes the following steps: Step 201: Obtain the latitude and longitude data of the social text location.

[0027] In some embodiments, the execution entity (e.g., a computing device) of the multi-feature fusion-based geolocation inference method for social text can obtain the longitude and latitude data of the location of the social text via a wired or wireless method. The location of the social text can be a pre-selected location. For example, the location of the social text can be a city. Thus, the longitude and latitude data corresponding to the city area can be obtained. Here, the longitude and latitude data can represent the user's location coordinates. The user's location coordinates can be uploaded simultaneously when the user sends the social text on the social platform.

[0028] It should be noted that the computing device described above can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed in the hardware devices listed above. It can be implemented as multiple software or software modules, for example, to provide distributed services, or as a single software or software module. No specific limitations are given here.

[0029] Step 202: geographically grid the latitude and longitude data to generate a geographical numbering grid.

[0030] In some embodiments, the execution entity may perform geographic grid division on the latitude and longitude data to generate a geographic number grid, wherein each grid in the geographic number grid corresponds to a grid number.

[0031] In some optional implementations of some embodiments, the execution entity performs geographic grid division on the latitude and longitude data to generate a geographic numbering grid, including: S1. Perform a compliance check on the latitude and longitude data to generate post-check latitude and longitude data. The post-check latitude and longitude data includes a sequence of latitude and longitude coordinates, each of which is within a preset geographic range. The compliance check can be used to remove latitude and longitude coordinates that are outside the preset geographic range.

[0032] As an example, the preset geographical range may be 3°52′ to 53°33′ north latitude and 73°40′ to 135°05′ east longitude.

[0033] S2, gridding the longitude and latitude intervals corresponding to the latitude and longitude data after the inspection according to a preset division size to obtain a divided geographic grid. Here, the longitude and latitude intervals can be the minimum circumscribed rectangular area of ​​the area where each longitude and latitude coordinate is located.

[0034] As an example, a geographic network can be divided into 0.5-degree intervals of longitude and latitude to obtain a divided geographic grid. This ensures the accuracy of geographic location inference while also balancing computational efficiency and practicality.

[0035] S3, using the longitude and latitude coordinate sequence included in the latitude and longitude data after the above check, determine the longitude index value and latitude index value corresponding to each grid in the above divided geographic grid. The longitude index value and latitude index value can be generated by the following formula: .

[0036] Among them, 73.55 is the decimal form of the country's minimum longitude, 73°33′E, and 3.85 is the decimal form of the country's minimum latitude, 3°51′N. The grid size is 0.5°×0.5°. The longitude of the location to be converted. is the latitude of the location to be converted. For the id The longitude index value of the grid. For the id The latitude index value of the grid. id Indicates the grid number.

[0037] S4, using the preset latitude priority coding condition, generating a grid number corresponding to each grid in the above-divided geographic grid. The grid number corresponding to each grid can be generated by the following formula: .

[0038] in, A grid number is assigned to each grid. Here, a latitude-first coding condition (strategy) is used to generate unique grid numbers. The grid numbers range from 0 to 12523, corresponding to 12524 grid cells. The latitude-first coding condition (strategy) can prioritize grid numbering based on latitude.

[0039] For example, see Figure 3 The diagram of geographic grid division is shown in Figure 1. Each grid corresponds to a unique grid number.

[0040] Optionally, you can also choose to return the grid number or generate the corresponding one-hot encoding according to your needs.

[0041] S5, determining the geographical grids after the division and the grid numbers corresponding to the grids as geographical number grids.

[0042] Step 203: construct a semantic feature vector based on the pre-acquired social text semantic information.

[0043] In some embodiments, the execution entity may construct a semantic feature vector based on pre-acquired social text semantic information.

[0044] In some optional implementations of some embodiments, the execution entity constructs a semantic feature vector based on pre-acquired social text semantic information, including: The first step is to deconstruct the semantic information of the social text to generate place name embedding vectors, place name context embedding vectors, and enhanced corpus vectors. The purpose of constructing semantic feature vectors is to mine the semantic information of social texts to realize the mining of explicit and implicit location information contained in the text.

[0045] The second step is to concatenate the place name embedding vector, the place name context embedding vector, and the enhanced corpus vector to obtain a semantic feature vector. The concatenation may be performed by vertically concatenating the place name embedding vector, the place name context embedding vector, and the enhanced corpus vector end to end to obtain the semantic feature vector.

[0046] Optionally, the execution entity deconstructs the social text semantic information to generate a place name embedding vector, a place name context embedding vector, and an enhanced corpus vector, including: S1, performing place name recognition on the social text semantic information to obtain a place name text word set. The place name recognition can be performed on the social text semantic information using a place name recognition algorithm to obtain the place name text word set.

[0047] As an example, the place name recognition algorithm may include but is not limited to at least one of the following: a bag-of-words model, a random conditional field algorithm, etc.

[0048] S2: Perform bidirectional context encoding on each place-name text word in the above place-name text word set to generate a place-name embedding vector. This can be done using the pre-trained BERT (Bidirectional Encoder Representations from Transformers) model to generate a 768-dimensional place-name entity embedding vector. This is used to capture the deep semantic relationships between social text words.

[0049] Optionally, to further enhance the model's ability to understand the semantics of place names in social text, the present invention designs a context capture mechanism based on a dynamic symmetric sliding window. This mechanism aims to comprehensively model multi-dimensional features such as the semantic associations and usage scenarios of place names, eliminating the ambiguity of entity reference. The calculation formula is as follows: .

[0050] In the above formula, is the basic window size corresponding to the place name entity type, among which the city type =8; District and county type =6; POI (Point of Interest) type =5;Structured address =7. t is the total length of the current social text, =50 is the maximum text length for statistics. =2 is the dynamic compensation factor. W The window size for calculation.

[0051] Therefore, the BERT pre-trained model is used to generate a context embedding vector for the obtained place name context.

[0052] In practice, due to the frequent occurrence of homonymous locations in social media texts, the diverse ways users express themselves, and the highly fragmented geographical information they convey, traditional corpora struggle to fully capture the semantic and spatial associations between geographic entities. By introducing a place name database and spatial semantic retrieval technology, we construct an augmented corpus K of locations that are semantically similar to the identified place names and spatially proximal to them, alleviating this problem.

[0053] Optionally, the execution subject deconstructs the social text semantic information to generate a place name embedding vector, a place name context embedding vector, and an enhanced corpus vector, further comprising: S1, collecting corpus data from the target social platform. The corpus data can be obtained from the target social platform through a preset collection code.

[0054] S2, extract geographic entities from the above corpus data to obtain a geographic entity label set.

[0055] As an example, in the corpus data, geographic entities are extracted from the original text using a place name entity recognition model, preliminarily identifying geographic entities such as city names, administrative divisions, POIs, and structured addresses. Here, the place name entity recognition model can be the aforementioned BERT model or a place name recognition algorithm.

[0056] S3: Verify the geographic entity tags in the geographic entity tag set to obtain a set of standardized place name tags. The geographic entity tag set can be sent to a verification terminal for manual correction to obtain standardized place name tags.

[0057] S4: Perform an address search on a preset place name database to generate a candidate address set. The address search is used to select candidate addresses that match standardized place name tags in the standardized place name tag set. The place name database can be a pre-selected database that includes all place names in a particular area.

[0058] As an example, Figure 4 As shown. For the identified place name set L Search through the place name database, and after retrieving the exact matching address, get each place name l A set of candidate locations C , the place names that cannot be retrieved are searched through the network interface (API (Application Programming Interface)) to build a candidate place set C .

[0059] S5, for each standardized place name tag in the above standardized place name tag set, perform the following steps: In the first step, vectors are constructed for the standardized place name labels and each candidate address in the corresponding candidate address group in the candidate address group set, resulting in a first place name vector and a second candidate vector group. The first place name vector and the second candidate vector are both 768-dimensional vectors. The BERT model can be used to construct vectors for the standardized place name labels and each candidate address in the corresponding candidate address group in the candidate address group set, resulting in a first place name vector and a second candidate vector group.

[0060] The second step is to determine the cosine similarity between the first place name vector and each second candidate vector in the second candidate vector group to obtain a cosine similarity group. The cosine similarity between the first place name vector and the second candidate vector can be determined by the following formula: .

[0061] in, is the cosine similarity between two place names (the first place name vector and the second candidate vector), and Represents the place names obtained by the BERT model (standardized place name labels) and the corresponding candidate place names (candidate addresses) BERT vector, and They represent the modulus of the place name vector respectively. . .

[0062] In the third step, based on the cosine similarity group, candidate addresses whose cosine similarity exceeds a preset distance threshold are eliminated from the candidate address group to obtain an eliminated address group.

[0063] As an example, a threshold of 0.5 is set to filter out low-quality candidate place names with semantic deviation exceeding 0.5. Figure 4 As shown, semantic similarity determination is to eliminate candidate addresses whose cosine similarity between the first place name vector and the second candidate vector exceeds a preset distance threshold (such as 0.5) through cosine similarity.

[0064] Optionally, during the spatial proximity determination phase, the Haversine distance formula is used to calculate the spherical distance between candidate locations with high semantic similarity and the original location (standardized place name label), and candidate locations with a spherical distance exceeding 20 km are eliminated. The formula is as follows: .

[0065] The latitude and longitude coordinates of the two locations are and , represents latitude, λ represents longitude,R is the radius of the earth, the inverse tangent function used here , the resulting y / x value is in the range [-π,π]. a represents the spherical distance.

[0066] In practice, through the dual constraint mechanism of semantic similarity and spatial proximity, a high-quality corpus with both semantic consistency and spatial relevance can be constructed. K (i.e., the address group after elimination).

[0067] As an example, Figure 4 As shown, in the spatial proximity determination stage, candidate addresses whose geographical proximity exceeds the preset proximity range (such as the spherical distance exceeds the preset spherical distance) can be proposed. This will result in a high-quality corpus collection. K .

[0068] S6: Generate an enhanced corpus vector based on the obtained removed address groups and the above-mentioned standardized place name tag set. The enhanced corpus vector can be obtained by performing deep semantic encoding on the obtained removed address groups and the above-mentioned standardized place name tag set using the above-mentioned BERT model.

[0069] As an example, refer to Figure 4 The enhanced corpus construction flowchart shown in Figure 2 is shown in Figure 2.

[0070] Step 204: construct a geographic feature vector based on the pre-acquired city population data.

[0071] In some embodiments, the execution entity may construct a geographic feature vector based on pre-acquired city population data.

[0072] In some optional implementations of some embodiments, the city population data is the population data of the social text location. The execution subject constructs a geographic feature vector based on the pre-acquired city population data, including: S1, mapping the above-mentioned urban population data to the above-mentioned geographical number grid to obtain a population distribution map. The urban population data can be allocated to corresponding grid cells according to the statistical results of districts and counties to obtain a population distribution map.

[0073] S2, smoothing the above population distribution map using a preset Gaussian kernel to obtain the target population distribution map.

[0074] As an example, the population distribution is smoothed by a 3×3 Gaussian kernel to alleviate the problems of excessively high local population density and sparse population distribution while retaining the spatial continuity characteristics of the population distribution.

[0075] S3. Create a geographic feature vector based on the target population distribution map. The dimension of the geographic feature vector is the same as the number of grid cells in the geographic numbering grid. Here, a geographic feature vector of corresponding dimension can be created based on the number of grid cells in the geographic numbering grid. The numerical value in the geographic feature vector represents the population of each grid cell in the target population distribution map.

[0076] Step 205: Input the semantic feature vector and the geographic feature vector into a geographic location reasoning model to obtain a geographic location reasoning result.

[0077] In some embodiments, the execution entity may input the semantic feature vector and the geographic feature vector into a geographic location inference model to obtain a geographic location inference result, wherein the geographic location inference result includes a grid number.

[0078] In some optional implementations of some embodiments, the geographic location inference model includes an input layer, a hidden layer, and an output layer. The execution entity inputs the semantic feature vector and the geographic feature vector into the geographic location inference model to obtain a geographic location inference result, including: S1: Input the semantic feature vectors into the input layer for linear transformation to obtain the transformed features. Here, the input layer of the geolocation inference model performs a linear transformation on the concatenated semantic feature vectors and introduces nonlinear mapping using the ReLU (Rectified Linear Unit) activation function to initially capture the underlying patterns in the data.

[0079] As an example, Figure 5 As shown, the model input layer receives the place name embedding vector , place name context embedding vector and enhanced corpus vectors Semantic feature vectors are constructed by splicing. Geographic feature vectors are not processed for the time being. , in order to preserve its integrity as prior knowledge alone.

[0080] S2: Input the transformed features into the hidden layer to obtain high-order semantic features, wherein the high-order semantic features have the same dimension as the geographic feature vector.

[0081] S3, superimpose the above high-order semantic features and the above geographic feature vector element by element to obtain deep fusion features.

[0082] As an example, Figure 5As shown, the hidden layer extracts more discriminative high-order semantic features through further linear transformation and nonlinear activation, resulting in the original logits. At this point, the logits and geographic features maintain the same dimensionality. They are then element-wise added together to achieve a deep fusion of semantic and spatial features.

[0083] S4: Input the deep fusion features into the output layer to generate a probability distribution vector with the same dimension as the number of grid cells in the geographic numbering network, and use the grid cell number corresponding to the dimension with the highest probability value in the probability distribution vector as the geographic location inference result. Each probability distribution value in the probability distribution vector corresponds to the probability value of a geographic grid cell.

[0084] As an example, Figure 5 As shown, the Softmax function is used to generate a probability distribution vector equal to the number of geographic grids, and each dimension corresponds to the probability value of a geographic grid. By selecting the dimension with the highest probability value (argmax), the dimension corresponds to the corresponding geographic grid number The geographic grid corresponding to this number is the final geographic location inference result of the model.

[0085] In practice, during the feature extraction and location prediction stages, the geolocation inference model with MLP (Multi-Layer Perceptron) as its core uses a layer-by-layer abstract feature extraction mechanism to learn high-order semantic representations and spatial distribution features from the original fusion vector, thereby achieving efficient and accurate geolocation inference.

[0086] Specifically, in order to achieve geographic location reasoning of text, the present invention formalizes this problem into a multi-class classification problem, and its mathematical expression is as follows: .

[0087] in, It represents a grid in the geographic number grid. Represents a text feature extraction function (for example, bag-of-words model, word frequency-inverse word frequency algorithm, etc.). It's a grid The weight of . It's a grid The bias term is used to adjust the classification decision boundary. For the text t Therefore, the geographic location inference is achieved by replacing the longitude and latitude coordinates of social text with the number of the geographic grid.

[0088] Compared with the existing technology, the key improvement of the present invention is that: the present invention takes explicit place name entities as the entry point, integrates explicit place name entities and implicit text semantics in the text, and realizes social text geographic location reasoning by constructing semantic feature vectors and geographic feature vectors of social text.

[0089] The above-described embodiments of the present disclosure have the following beneficial effects: The multi-feature fusion-based geolocation inference method for social text, as described in some embodiments of the present disclosure, can improve inference accuracy at different scales and address the geolocation inference problem in social text. Given that deep learning models can achieve a good understanding of social text and fully capture multiple aspects of text features, which is highly suitable for the geolocation inference task in social text, the present invention proposes a multi-feature fusion geolocation inference method for social text to address the geolocation inference problem in social text. This method utilizes a multi-layered feature extraction and integration mechanism, starting with the disambiguation of place-name entities and understanding text semantics, to infer the actual geolocation referenced by the text. The actual geographic space is then divided into a geographic network based on latitude and longitude intervals, ensuring both accuracy and computational efficiency. To achieve geolocation inference in social text, it is necessary to exploit as many features as possible in the existing data. The features mined through feature engineering primarily include semantic and geographic features. Finally, the geolocation inference model transforms the original fusion vector through a layer-by-layer feature extraction mechanism to learn high-level semantic representations and spatial distribution features, thereby achieving efficient and accurate geolocation inference. In summary, the multi-feature fusion geolocation inference framework for social text proposed in this paper enables multi-level semantic mining of social text and enhancement of geographic features, enabling better geolocation inference for social text. It achieves good location inference results across the collected corpus. It also achieves a certain degree of comprehensive utilization of explicit and implicit semantic features in social text. This improves inference accuracy at different scales and addresses the geolocation inference problem for social text.

[0090] Further references Figure 6 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a social text geographic location inference device based on multi-feature fusion. These device embodiments are similar to Figure 2 Corresponding to the method embodiments shown, the social text geographic location inference device based on multi-feature fusion can be specifically applied to various electronic devices.

[0091] like Figure 6As shown, in some embodiments, a device 600 for inferring a social text location based on multi-feature fusion includes: an acquisition unit 601, a grid division unit 602, a first construction unit 603, a second construction unit 604, and an input unit 605. The acquisition unit 601 is configured to acquire the longitude and latitude data of the social text location; the grid division unit 602 is configured to perform geographic grid division on the longitude and latitude data to generate a geographic number grid, wherein each grid in the geographic number grid corresponds to a grid number; the first construction unit 603 is configured to construct a semantic feature vector based on pre-acquired social text semantic information; the second construction unit 604 is configured to construct a geographic feature vector based on pre-acquired city population data; and the input unit 605 is configured to input the semantic feature vector and the geographic feature vector into a geographic location inference model to obtain a geographic location inference result, wherein the geographic location inference result includes a grid number.

[0092] It is understandable that the units recorded in the social text location inference device 600 based on multi-feature fusion are similar to the reference Figure 2 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the social text geographic location inference device 600 based on multi-feature fusion and the units contained therein, and will not be repeated here.

[0093] Reference below Figure 7 , which shows a schematic structural diagram of an electronic device (such as a computing device) suitable for implementing some embodiments of the present disclosure. Figure 7 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure. Figure 7 As shown, the computer device includes a processor, a memory and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory. The non-volatile storage medium may store an operating system and a computer program. The computer program includes program instructions, which, when executed, may enable the processor to execute any of the above methods. The processor is used to provide computing and control capabilities to support the operation of the entire computer device. The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium, which, when executed by the processor, may enable the processor to execute any of the above methods. The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 7 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present disclosure, and does not constitute a limitation on the computer device to which the solution of the present disclosure is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0094] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0095] In one embodiment, the processor is used to run a computer program stored in a memory to implement the following steps: obtaining the latitude and longitude data of the social text location; performing geographic grid division on the latitude and longitude data to generate a geographic number grid, wherein each grid in the geographic number grid corresponds to a grid number; constructing a semantic feature vector based on pre-acquired social text semantic information; constructing a geographic feature vector based on pre-acquired city population data; inputting the semantic feature vector and the geographic feature vector into a geographic location reasoning model to obtain a geographic location reasoning result, wherein the geographic location reasoning result includes a grid number.

[0096] An embodiment of the present disclosure further provides a computer-readable storage medium, on which a computer program is stored. The computer program includes program instructions. The method implemented when the program instructions are executed can refer to the various embodiments of the method described above in the present disclosure.

[0097] The computer-readable storage medium may be an internal storage unit of the computer device described in the aforementioned embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a flash memory card, etc., provided on the computer device.

[0098] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0099] The above descriptions are merely some preferred embodiments of the present disclosure and illustrate the underlying technical principles. Those skilled in the art should understand that the scope of the invention encompassed by the embodiments of the present disclosure is not limited to technical solutions formed by specific combinations of the aforementioned technical features. It also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents, without departing from the aforementioned inventive concept. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A social text location inference method based on multi-feature fusion, characterized by: include: Get the latitude and longitude data of the social text location; Performing geographic grid division on the latitude and longitude data to generate a geographic number grid, wherein each grid in the geographic number grid corresponds to a grid number; Constructing a semantic feature vector based on pre-acquired social text semantic information; Construct geographic feature vectors based on pre-acquired urban population data; The semantic feature vector and the geographic feature vector are input into a geographic location reasoning model to obtain a geographic location reasoning result, wherein the geographic location reasoning result includes a grid number.

2. The method according to claim 1, characterized in that The step of dividing the latitude and longitude data into geographic grids to generate a geographic numbering grid includes: Performing a compliance check on the latitude and longitude data to generate checked latitude and longitude data, wherein the checked latitude and longitude data includes a latitude and longitude coordinate sequence, and each latitude and longitude coordinate in the latitude and longitude coordinate sequence is within a preset geographical range; Dividing the longitude and latitude intervals corresponding to the latitude and longitude data after the inspection into grids according to a preset division size to obtain a divided geographic grid; Determine the longitude index value and latitude index value corresponding to each grid in the divided geographic grid using the longitude and latitude coordinate sequence included in the checked longitude and longitude data; Using a preset latitude-first coding condition, generating a grid number corresponding to each grid in the divided geographic grid; The divided geographical grids and the grid numbers corresponding to the respective grids are determined as geographical number grids.

3. The method according to claim 1, characterized in that The step of constructing a semantic feature vector based on pre-acquired social text semantic information includes: Deconstructing the social text semantic information to generate a place name embedding vector, a place name context embedding vector, and an enhanced corpus vector; The place name embedding vector, the place name context embedding vector and the enhanced corpus vector are concatenated to obtain a semantic feature vector.

4. The method according to claim 3, characterized in that The city population data is the population data of the social text location, wherein constructing the geographic feature vector based on the pre-acquired city population data includes: Mapping the city population data to the geographic number grid to obtain a population distribution map; Smoothing the population distribution map using a preset Gaussian kernel to obtain a target population distribution map; A geographic feature vector is established according to the target population distribution map, wherein the dimension of the geographic feature vector is the same as the number of grids in the geographic numbering grid.

5. The method according to claim 4, characterized in that The geographic location inference model includes: an input layer, a hidden layer, and an output layer, wherein the semantic feature vector and the geographic feature vector are input into the geographic location inference model to obtain a geographic location inference result, including: Inputting the semantic feature vector into the input layer for linear transformation to obtain transformed features; Inputting the transformed features into the hidden layer to obtain high-order semantic features, wherein the high-order semantic features have the same dimension as the geographic feature vector; Superimposing the high-order semantic features and the geographic feature vector element by element to obtain a deep fusion feature; The deep fusion feature is input into the output layer to generate a probability distribution vector with the same dimension as the number of grids in the geographic numbering network, and the grid number corresponding to the dimension with the highest probability value in the probability distribution vector is used as the geographic location inference result, wherein each probability distribution value in the probability distribution vector corresponds to the probability value of a geographic grid.

6. The method according to claim 3, characterized in that Deconstructing the social text semantic information to generate a place name embedding vector, a place name context embedding vector, and an enhanced corpus vector includes: Performing place name recognition on the social text semantic information to obtain a place name text word set; Perform bidirectional context encoding on each place name text word in the place name text word set to generate a place name embedding vector.

7. The method according to claim 6, characterized in that The deconstructing the social text semantic information to generate a place name embedding vector, a place name context embedding vector, and an enhanced corpus vector further includes: Collect corpus data of the target social platform; Extracting geographic entities from the corpus data to obtain a geographic entity tag set; Verifying the geographic entity tags in the geographic entity tag set to obtain a standardized place name tag set; Performing an address search on a preset place name database to generate a candidate address group set, wherein the address search is used to select candidate addresses that match the standardized place name labels in the standardized place name label set; For each standardized place name tag in the set of standardized place name tags, perform the following steps: Performing vector construction on each candidate address in the corresponding candidate address group in the standardized place name label and the candidate address group set to obtain a first place name vector and a second candidate vector group, wherein the first place name vector and the second candidate vector are both 768-dimensional vectors; determining a cosine similarity between the first place-name vector and each second candidate vector in the second candidate vector group to obtain a cosine similarity group; Eliminating candidate addresses whose corresponding cosine similarities exceed a preset distance threshold from the candidate address group according to the cosine similarity group to obtain an eliminated address group; An enhanced corpus vector is generated based on the obtained eliminated address groups and the standardized place name tag set.

8. A social text geographic location inference device based on multi-feature fusion, characterized in that: include: an acquisition unit configured to acquire latitude and longitude data of a social text location; a grid division unit configured to perform geographic grid division on the latitude and longitude data to generate a geographic number grid, wherein each grid in the geographic number grid corresponds to a grid number; A first construction unit is configured to construct a semantic feature vector based on pre-acquired social text semantic information; A second construction unit is configured to construct a geographic feature vector based on pre-acquired urban population data; The input unit is configured to input the semantic feature vector and the geographic feature vector into a geographic location reasoning model to obtain a geographic location reasoning result, wherein the geographic location reasoning result includes a grid number.

9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable medium, characterized in that A computer program is stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images