Location search method, electronic device, and storage medium

Abstract

Embodiments of the present application disclose a place retrieval method, an electronic device and a storage medium, comprising: obtaining a to-be-retrieved text; performing entity segmentation on the to-be-retrieved text to obtain at least one entity corresponding to the to-be-retrieved text; determining an entity identifier corresponding to each entity in the to-be-retrieved text according to a preset association relationship; and outputting a target place corresponding to the to-be-retrieved text based on the entity identifier corresponding to each entity and an identifier type corresponding to the entity identifier. The scheme can not only improve the recalled places during place retrieval, but also improve the accuracy of place retrieval.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a place retrieval method, an electronic device and a storage medium. BACKGROUND

[0002] With the development of intelligent devices and mobile internet technology, new travel methods combined with the Internet have also developed vigorously. In order to better meet the needs of users, accurately identifying the destination that the user wants to go to is crucial for place retrieval.

[0003] The current place retrieval scheme mainly recommends the destination for the place word input by the user, so that the user can select the destination. The basis for selection is mainly based on the scoring of the existing results in the database, and the result with the highest similarity score is returned. However, this method needs to rely on a large and complete database, and if the address searched by the user is not in the database, it cannot be predicted. Secondly, the result is limited by the effect of the sorting algorithm, and the text structure similarity may be greater than the semantic similarity. It can be seen that in the current place retrieval scheme, not only the recall rate is low, but also the accuracy is poor. SUMMARY

[0004] The embodiments of the present application provide a place retrieval method, an electronic device and a storage medium, which can not only improve the recalled place during place retrieval, but also improve the accuracy of place retrieval.

[0005] The embodiments of the present application provide a place retrieval method, comprising:

[0006] Obtaining a to-be-retrieved text;

[0007] Performing entity segmentation on the to-be-retrieved text to obtain at least one entity corresponding to the to-be-retrieved text;

[0008] According to a preset association relationship, determining an entity identifier corresponding to each entity in the to-be-retrieved text;

[0009] Based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier, outputting a target place corresponding to the to-be-retrieved text.

[0010] Optionally, in some embodiments, the outputting of the target place corresponding to the to-be-retrieved text based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier comprises:

[0011] Identifying the identifier type corresponding to the entity identifier;

[0012] Based on the identifier type, determining a place retrieval strategy corresponding to the to-be-retrieved text;

[0013] According to the determined place search strategy and the entity identifier corresponding to each entity, a target place corresponding to the text to be searched is output.

[0014] Optionally, in some embodiments, the outputting of the target place corresponding to the text to be searched according to the determined place search strategy and the entity identifier corresponding to each entity comprises:

[0015] An administrative region corresponding to the text to be searched is obtained.

[0016] A first target entity is determined from the entities based on a first place search strategy.

[0017] A target place corresponding to the text to be searched is output based on the administrative region and the entity identifier corresponding to the first target entity.

[0018] Optionally, in some embodiments, the outputting of the target place corresponding to the text to be searched based on the administrative region and the entity identifier corresponding to the first target entity comprises:

[0019] A first recall text corresponding to the text to be searched is determined from a preset text library based on the administrative region and the entity identifier corresponding to the first target entity.

[0020] The first recall text satisfying a preset condition is determined as a first target text, and a place corresponding to the first target text is determined as the target place corresponding to the text to be searched.

[0021] Optionally, in some embodiments, the outputting of the target place corresponding to the text to be searched according to the determined place search strategy and the entity identifier corresponding to each entity comprises:

[0022] A second recall text corresponding to the text to be searched is constructed based on a second place search strategy.

[0023] An edit distance similarity between the text to be searched and each second recall text is calculated.

[0024] A second recall text with an edit distance similarity greater than a preset value is determined as a candidate text.

[0025] A candidate text with the greatest edit distance similarity is determined as a second target text, and a place corresponding to the second target text is determined as the target place corresponding to the text to be searched.

[0026] Optionally, in some embodiments, the method further comprises:

[0027] When a prediction result of the second place search strategy is an empty set, a nearby point prediction is performed on the entities.

[0028] Output the target location corresponding to the text to be searched based on the prediction result.

[0029] Optionally, in some embodiments, the method further comprises:

[0030] When the prediction result of the nearby point is an empty set, retain the first target entity in the text to be searched;

[0031] Output the target location corresponding to the text to be searched according to the first target entity.

[0032] Optionally, in some embodiments, before the step of outputting the target location corresponding to the text to be searched based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier, the method further comprises:

[0033] Obtain a current location search environment;

[0034] Retain a third target entity in the text to be searched based on the current location search strategy;

[0035] The step of outputting the target location corresponding to the text to be searched based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier comprises the step of outputting the target location corresponding to the text to be searched based on the entity identifier corresponding to the third target entity and the identifier type corresponding to the entity identifier.

[0036] Correspondingly, the application also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the steps of any of the above methods.

[0037] The application also provides a computer-readable storage medium storing a computer program, wherein the computer program is executable on a processor to implement the steps of any of the above methods.

[0038] The location search scheme provided by the embodiments of the application comprises the following steps: after obtaining a text to be searched, performing entity segmentation on the text to be searched to obtain at least one entity corresponding to the text to be searched; then, determining an entity identifier corresponding to each entity in the text to be searched according to a preset association relationship; and finally, outputting a target location corresponding to the text to be searched based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier. In the location search scheme provided by the application, the text to be searched can be subjected to entity segmentation to reduce the influence of meaningless words on the search result, and the target location corresponding to the text to be searched is output based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier, so that not only the recalled locations during location search can be improved, but also the accuracy of location search can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0040] Figure 1 is a flowchart of a place retrieval method provided by an embodiment of the present application;

[0041] Figure 2 is a schematic diagram of address structuring in the place retrieval method provided by an embodiment of the present application;

[0042] Figure 3 is a structural schematic diagram of a place retrieval device provided by an embodiment of the present application;

[0043] Figure 4 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort belong to the scope of protection of the present application.

[0045] An embodiment of the present application provides a place retrieval method, device, electronic device and storage medium.

[0046] The place retrieval device can be integrated in a server or a terminal. The server can include a standalone server or a distributed server, or a server cluster composed of multiple servers. The terminal can include a mobile phone, a tablet computer or a personal computer (PC).

[0047] The following will be described in detail. It should be noted that the description order of the following embodiments is not used to limit the priority order of the embodiments.

[0048] A place retrieval method includes: obtaining a to-be-retrieved text; performing entity segmentation on the to-be-retrieved text to obtain at least one entity corresponding to the to-be-retrieved text; determining entity identifiers corresponding to each entity in the to-be-retrieved text according to a preset association relationship; and outputting a target place corresponding to the to-be-retrieved text based on the entity identifiers corresponding to each entity and the identifier types corresponding to the entity identifiers.

[0049] Please refer toFigure 1 , Figure 1 The flowchart of the place retrieval method provided by the embodiment of the present application is shown in the figure. The specific flow of the place retrieval method can be as follows:

[0050] 101, obtaining a text to be retrieved.

[0051] The text to be retrieved can be a text input by a user, can also be a text saved in a local database, or can be a text obtained by accessing a network interface, which is determined according to actual conditions.

[0052] For example, a user inputs "XX district XX road" in a car-hailing software, and then the "XX district XX road" is the text to be retrieved. For another example, a user inputs "XX community" in a freight software, and then the "XX community" is the text to be retrieved.

[0053] 102, performing entity segmentation on the text to be retrieved to obtain at least one entity corresponding to the text to be retrieved.

[0054] The entity is also referred to as an entity word, and the entity word refers to a word with a specific semantic meaning, which can be a noun capable of referring to a certain thing, such as Zhang San or XX community, etc.

[0055] For example, specifically, a preset named entity recognition (NER) algorithm can be used to perform entity segmentation on the text to be retrieved to obtain at least one entity corresponding to the text to be retrieved. NER is a very basic task in natural language processing, and is an important basic tool for numerous NLP tasks such as information extraction, question answering system, syntax analysis, machine translation, etc. A named entity generally refers to an entity in a text with a specific meaning or strong reference, which is generally divided into three categories of entity classes, time classes, and number classes, and seven subcategories of personal names, place names, organization names, time, date, currency, and percentage. NER is to extract the above entities from unstructured input text, and can identify more categories of entities according to business requirements.

[0056] The named entity recognition algorithm can include a dictionary and rule-based method, a traditional machine learning method, and a deep learning-based method.

[0057] The rule-based NER system relies on manually developed rules. The design of rules is generally based on syntax, grammar, lexical patterns, and knowledge in specific fields, etc. The dictionary is composed of a feature word dictionary and an external dictionary, and the external dictionary refers to an existing common sense dictionary. After the rules and the dictionary are developed, the matching method is usually used to process the text to achieve named entity recognition.

[0058] In the machine learning based method, the named entity recognition is taken as a sequence labeling problem. Compared with the classification problem, the current predicted label in the sequence labeling problem is not only related to the current input feature, but also related to the previous predicted label, that is, there is a strong interdependence between the predicted label sequence. The traditional machine learning methods adopted mainly include: hidden Markov model, maximum entropy, maximum entropy Markov model, support vector machine and conditional random field.

[0059] The reasons why the NER uses deep learning mainly include: 1. The NER is suitable for nonlinear conversion. 2. The deep learning avoids the construction of a large number of artificial features, and saves a lot of efforts in designing the NER function. 3. The deep learning can be trained through gradient propagation, so that a more complex network can be constructed. 5. The end-to-end training method.

[0060] In the present application, the corresponding NER algorithm can be selected according to the actual needs, and will not be repeated here.

[0061] 103. According to the preset association relationship, the entity identifiers corresponding to each entity in the text to be searched are determined.

[0062] In the present application, the address structure of the text to be searched is structured, that is, after the entity segmentation of the text to be searched, each entity is given a corresponding entity identifier, please refer to Figure 2 The address structure method divides the address into province, city, district, township, road, intersection, orientation, distance and the like. The actual processing data format is a list composed of (component name, start position, end position, type, level identifier) elements, for example: [(“Jiangsu Province”, 0, 3, “provice”, 2), …, (“Lu Xi”, x, x, “orientation”, 28)], as shown in Table 1:

[0063]

[0064]

[0065] Table 1

[0066] Among them, the association relationship between the level, the place name and the commonly used keywords can be constructed in advance, the keywords mentioned in the present application are only examples and do not limit the present application, the association relationship of the present application can also be the relationship between other attributes, which can be selected according to the actual situation, and will not be repeated here.

[0067] 104. Based on the entity identifiers corresponding to each entity and the identifier types corresponding to the entity identifiers, the target place corresponding to the text to be searched is output.

[0068] For example, specifically, for the text to be retrieved "Beijing City, Changping District, Innovation Road, No. 11, 10th Floor, Room 1015", it is known from Table 1 that the corresponding identifier is "2.4.8.21.23.24.25", and thus the target location corresponding to the text to be retrieved can be obtained.

[0069] Further, in the actual retrieval process, the elements contained in the text to be retrieved are not comprehensive, for example, the text to be retrieved A is "Changping District, Innovation Road, No. 11", and the text to be retrieved B is "Beijing City, Changping District, Innovation Road", that is, the entities contained in the text to be retrieved are different, and thus the corresponding location retrieval strategy needs to be determined according to the entities contained in the text to be retrieved, that is, in some embodiments, the step of "outputting the target location corresponding to the text to be retrieved based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier" comprises:

[0070] (11) identifying the identifier type corresponding to the entity identifier;

[0071] (12) determining the location retrieval strategy corresponding to the text to be retrieved based on the identifier type;

[0072] (13) outputting the target location corresponding to the text to be retrieved according to the determined location retrieval strategy and the entity identifier corresponding to each entity.

[0073] For example, specifically, the application predefines the key components (i.e. the identifier type of the target entity) in order to subsequently determine the corresponding location retrieval strategy, and at the same time, the redundant components in the text to be retrieved can be removed, thereby improving the recall rate of the location.

[0074] Please refer to Table 2, which is the key component table predefined by the application

[0075] Basic place name Level Primary sub-point type Community, village 6 - Road 8 Primary point Point of interest 10 Primary point Road house number 21 Sub-point Building number 22 Sub-point Place number 26 Sub-point

[0076] Table 2

[0077] Specifically, the address structured text to be retrieved is traversed to determine whether the entity identifier of the target entity in the text to be retrieved is in the key component table, and if so, the entity label (entity identifier) is extracted, otherwise it is skipped. For example, "Beijing City, Changping District, Innovation Road, No. 11, 10th Floor, Room 1015" hits the key component list [8, 21], that is, ["Innovation Road", "No. 11"].

[0078] Optionally, in some embodiments, the main point can be determined as the first target entity, the sub-point can be determined as the second target entity, the main point recall is to identify the main point component from the address component list on the basis of address structuring (i.e. the entity segmented to-be-retrieved text), and the main point component is used to query the related point in the database. When the to-be-retrieved text only contains the main point, firstly, the administrative region corresponding to the to-be-retrieved text needs to be determined, and then the target location corresponding to the to-be-retrieved text is output based on the administrative region and the entity identifier. That is, optionally, in some embodiments, the step of "outputting the target location corresponding to the to-be-retrieved text according to the determined location retrieval strategy and the entity identifier corresponding to each entity" can specifically include:

[0079] (21) obtaining the administrative region corresponding to the to-be-retrieved text;

[0080] (22) determining the first target entity among the entities based on the first location retrieval strategy;

[0081] (23) outputting the target location corresponding to the to-be-retrieved text based on the administrative region and the entity identifier corresponding to the first target entity.

[0082] There are two ways to obtain the administrative region, the first way is to identify the city by using address structured information, and the second way is to use the city selected by the user. Optionally, in some embodiments, when the city is contained in the to-be-retrieved text input by the user and the city is additionally selected by the user, the city contained in the to-be-retrieved text input by the user is used as the criterion.

[0083] Further, when it is identified that the to-be-retrieved text only contains the main point (i.e. the first target entity), the recall text corresponding to the to-be-retrieved text is queried in the preset text library, and the corresponding target location is output based on the query obtained recall text. That is, optionally, in some embodiments, the step of "outputting the target location corresponding to the to-be-retrieved text based on the administrative region and the entity identifier corresponding to the first target entity" includes:

[0084] (31) determining the first recall text corresponding to the to-be-retrieved text in the preset text library based on the administrative region and the entity identifier corresponding to the first target entity;

[0085] (32) determining the first target text that meets the preset condition as the first target text, and determining the location corresponding to the first target text as the target location corresponding to the to-be-retrieved text.

[0086] For example, specifically, as shown in Table 2, in some embodiments of the present application, the point of interest (POI) and the road are defined as the main point (i.e. the first target entity), and then the preset text library is searched based on the main point component contained in the to-be-retrieved text. The search strategy is as follows:

[0087] a) When roads and POIs exist in the text to be retrieved, first find data in the database that meets the conditions of roads and POIs at the same time; if no data is found, return the data of roads that meet the conditions in the preset text library; if still not found, do not return data;

[0088] b) When only roads exist in the text to be retrieved, find data of roads that meet the conditions in the preset text library;

[0089] c) When only POIs exist in the text to be retrieved, find data of POIs that meet the conditions in the preset text library.

[0090] The specific matching strategy is as follows:

[0091] First, the keyword screening of the text to be retrieved is performed, and the keyword screening refers to filtering the recall list according to the key components hit in the text to be retrieved. If the recall lacks the component level label corresponding to the text to be retrieved, the recall text is retained, otherwise it is excluded.

[0092] For example, the text to be retrieved is "Beijing Changping District, Innovation Road No. 11", the entity identifier is [2, 4, 8, 21], the recall text is "Beijing Changping District, Innovation Road No. 13", and the entity identifier is [2, 4, 8, 21] and [2, 4, 8], respectively. The result obtained after filtering is "Beijing Changping District, Innovation Road No. 13". Then, keyword matching is performed,

[0093] The keyword matching refers to checking whether the corresponding component content in the text to be retrieved and the corresponding component content in the recall text match according to the pre-defined address components. If all of them match, they are retained, and if there is any unmatched component, they are excluded. The specific steps can be as follows:

[0094] Step a) check which pre-defined entity identifiers in the text to be retrieved hit the pre-defined identifiers;

[0095] Step b) if the corresponding pre-defined identifier exists in the recall text, check whether the components corresponding to the same identifier in the text to be retrieved and the recall text all match;

[0096] There are two matching strategies, prefix matching and similarity matching:

[0097] 1) Prefix matching: the component content of the recall text is prefixed by the component content of the text to be retrieved;

[0098] 2) Similarity matching: calculate the edit distance similarity between the component content of the recall text and the component content of the text to be retrieved, and the similarity meets a certain threshold;

[0099] The edit distance is a quantitative measure of the difference between two strings (e.g., English words), which measures how many operations are needed to change one string into another. The edit distance can be used in natural language processing, for example, spelling checking can determine which (or which) is more likely to be the correct word based on the edit distance between the misspelled word and other correct words.

[0100] For example, the text to be searched: "No. 11 Innovation Road", recall text A: "No. 13 Innovation Road", recall text B: "No. 15 Union Road", recall text C: "Innovation Road Science and Technology Building" After filtering, the result is: recall text: "No. 13 Innovation Road". Among them, the road corresponding to the recall text B does not match, and the recall text C appears in the POI that does not exist in the text to be searched.

[0101] Further, the edit distance similarity between the text to be searched and the recall text is calculated, and the recall text with an edit distance similarity less than a preset value s1 is removed, thereby filtering the point of the recall text. Subsequently, in the retained recall text, the recall text with the largest edit distance similarity (hereinafter referred to as A-type text) is selected, and when there are two or more A-type texts, the shortest A-type text is selected. The purpose of this is to ensure that the address content of the output recall text has as little redundant information as possible. Finally, the confidence of the recall text is calculated, which can be measured by the text similarity between the recall text and the text to be searched. For example, the recall text with a text similarity greater than or equal to a preset value s2 is determined as the first target text, and the target place corresponding to the first target text is output.

[0102] Optionally, in some embodiments, when the text to be searched contains a main point and a sub-point, the edit distance similarity between the text to be searched and the recall text can be calculated, and then the target place is determined based on the edit distance similarity. That is, in some embodiments, the step "output the target place corresponding to the text to be searched according to the determined place retrieval strategy and the entity identifier corresponding to each entity" can specifically include:

[0103] (41) constructing a second recall text corresponding to the text to be searched based on a second place retrieval strategy;

[0104] (42) calculating the edit distance similarity between the text to be searched and each second recall text;

[0105] (43) determining the second recall text with an edit distance similarity greater than a preset value as a candidate text;

[0106] (44) determining the candidate text with the largest edit distance similarity as a second target text, and determining the place corresponding to the second target text as the target place corresponding to the text to be searched.

[0107] For example, specifically, the main point and the sub-point corresponding to the second retrieval strategy are determined, then the second recall text is constructed according to the determined main point and sub-point, then the edit distance similarity between the text to be retrieved and each second recall text is calculated, and then the second recall text with an edit distance similarity greater than a preset value is determined as a candidate text, and when the second target text can be determined by the strategy, the target place corresponding to the second target text is output, and the specific process can refer to the process of the main point prediction.

[0108] It should be noted that when the prediction result of the second retrieval strategy is an empty set, i.e., the target place cannot be output under the second retrieval strategy, the adjacent point prediction method can be performed, i.e., in some embodiments, the place retrieval method of the application can further include:

[0109] (51) When the prediction result of the second place retrieval strategy is an empty set, the adjacent point of each entity is predicted;

[0110] (52) Based on the prediction result, the target place corresponding to the text to be retrieved is output.

[0111] The application provides the following adjacent point prediction methods, which can be as follows:

[0112] Method one:

[0113] Step 1. Second recall text component summary, the component content in the second recall text list is summarized, and a dictionary key-value pair is formed according to the level label, the level label is the key, and the component content is the value. And meet the condition that the same component content under the same label in the second recall text is only retained 1, that is, the value corresponding to one key is not repeated;

[0114] Example:

[0115] {“8”:[innovation road],“21”:[11, 13, 19]}

[0116] Step 2. Determine the key component to be predicted, based on the dictionary generated in step 1, the component label to be predicted is inferred.

[0117] The inference logic is:

[0118] a) The sub-point component is the component corresponding to 21, 22 and 26 in Table 2 in section 1.2;

[0119] b) The component content corresponding to the component label to be predicted is the most;

[0120] c) In the case where the number of values corresponding to different labels is the same, the higher the label level is, the better;

[0121] Example: The search label to be inferred is “21”

[0122] Step 3. Numerical mapping, convert the number components in step 1 to be searched into integer type values for subsequent search of adjacent numbers. This step is processed at the same time for the text to be retrieved and the second recall text. In the number list corresponding to the second recall text label, search for the nearest point from the number of the text to be retrieved.

[0123] The numerical mapping logic is divided into two types: numerical type and alphabetical type. The specific scheme is as follows:

[0124] a) Remove the suffix in the number component;

[0125] b) Determine the number type, whether it is a number or a letter;

[0126] c) Numerical conversion;

[0127] Example:

[0128] [11, 13, 19] is converted to [11, 13, 19]

[0129] Step 4. Nearest point search, first sort the numerical numbers, then search for the number with the smallest difference from the number of the text to be retrieved.

[0130] The specific scheme is as follows:

[0131] a) Sort the numerical numbers corresponding to the second recall text label in ascending order to facilitate subsequent number search;

[0132] b) Use binary search method to find the number with the smallest difference between the number of the second recall text and the number of the text to be retrieved;

[0133] Example:

[0134] The number of the text to be retrieved is 14, and the second recall text list is [11, 13, 19], so the nearest point number searched is 13.

[0135] Step 5. Nearest neighbor recall, based on the sorted number list of the second recall text and the found nearest point, find the adjacent candidate point, and narrow down the candidate set for the subsequent point prediction strategy.

[0136] The adjacent candidate point screening strategy is:

[0137] a) Number proximity: refers to the adjacent numbers selected from the second recall text number list according to the rule that the difference between the number of the text to be retrieved and the number is less than y numbers;

[0138] b) Index proximity: refers to the adjacent numbers selected from the second recall text number list according to the rule that the number found by the nearest point search method is within z positions before and after the index position;

[0139] Example:

[0140] The text number to be retrieved is 14, and the second recall text number list is [11, 13, 19]

[0141] The number proximity threshold is 2, and 13 is filtered out;

[0142] The index proximity threshold is 2, and 11, 13, and 19 are filtered out;

[0143] Step 6. Point prediction strategy, calculate the proximity point of step 5 according to the pre-set strategy, and predict the return result. There are five kinds of sub-point prediction strategies, and the calculation logic of each strategy is different, as follows:

[0144] a) Sub-point number hit strategy

[0145] The sub-point number hit strategy refers to when the second recall text sub-point component list appears exactly in the sub-point component of the text to be retrieved, it is judged whether the second recall text sub-point can be returned as the return result, if it meets the similarity threshold, it is returned, otherwise it is not returned.

[0146] 1) Find the second recall text sub-point component consistent with the sub-point component of the text to be retrieved, and possibly unify the sub-point number corresponding to multiple second recall text data;

[0147] 2) Judge whether the similarity exceeds the specified threshold, and eliminate the results that do not meet the threshold;

[0148] 3) If there is no second recall text data that meets the threshold, this strategy returns empty;

[0149] 4) If there is second recall text data that meets the threshold, this strategy returns any second recall text data as the most red result;

[0150] 5) Calculate the confidence, and the confidence is equal to the similarity.

[0151] Example:

[0152] The text number to be retrieved is 13, and the second recall text number list is [11, 13, 19]

[0153] The strategy hits the sub-point number 13

[0154] Method two:

[0155] The road double-side number generation strategy refers to when the main point component in the text to be retrieved is a road (level 8) and the sub-point number component is a house number (level 21), the second recall text can find the numbers adjacent to the sub-point number of the text to be retrieved. In this case, a virtual point is generated as the return result through the two side numbers.

[0156] 1) Check if the main point of the text to be searched is of road type, if yes, continue, otherwise end;

[0157] 2) Determine the parity of the value of the sub-point of the text to be searched;

[0158] 3) According to the parity of the value of the sub-point of the text to be searched, filter the sub-points in the second recall text list that are inconsistent with the parity of the sub-point of the text to be searched;

[0159] 4) Filter the second recall text sub-point and the sub-point number difference of the text to be searched, if there is only a single adjacent point at this time, end, otherwise continue;

[0160] 5) Select the two second recall text sub-point numbers with the smallest distance from the sub-point number of the text to be searched, and find the second recall text data corresponding to the number;

[0161] 6) If the data corresponding to the left number (or right number) is greater than or equal to 3, the abnormal point needs to be removed. The specific method is to calculate the center point of the second recall text data longitude and latitude, and filter the point with large distance from the center;

[0162] 7) Calculate the average value of the left number (or right number) using the filtered point to obtain the final longitude and latitude coordinates corresponding to the number;

[0163] Based on the final longitude and latitude coordinates of the left and right adjacent numbers, the longitude and latitude of the middle number (the sub-point number of the text to be searched) are predicted by linear interpolation method. The specific formula is as follows

[0164]

[0165]

[0166] Where, lon pred is the predicted longitude coordinate, lon right is the longitude coordinate of the right side, lon left is the longitude coordinate of the left side, num pred is the predicted number, num left is the number of the left side, lat pred is the predicted latitude coordinate, lat right is the latitude coordinate of the right side, lat left is the latitude coordinate of the left side.

[0167] Calculate the confidence, and the confidence is obtained by weighted sum of similarity and distance, the specific formula is as follows

[0168]

[0169] confidence = similarity + (1 - 2 * distance / threshold) * 0.05 (4)

[0170] Wherein, geodesic represents the spherical coordinate system distance calculation formula, the input two point latitude and longitude can get the distance; num represents the number, distance is the distance divided by the number difference, which represents the distance of the difference of the unit number difference, which is used to normalize the distance difference; threshold is the set distance threshold; the confidence formula is the sum of the text similarity similarity and the distance distance weight.

[0171] Example:

[0172] The text number to be searched is 13, and the second recall text number list is [11, 15, 19]

[0173] 11 and 15 can be found in the second recall text, so the corresponding latitude and longitude of the number 13 is calculated by using the generation strategy, and the virtual point generated by the latitude and longitude and the number is returned as the return result.

[0174] Method three:

[0175] The one-sided number generation strategy refers to when the main point component in the text to be searched is a road (level 8) and the predicted sub-point number component is a house number (level 21), the second recall text can find the number adjacent to the sub-point number of the text to be searched on one side. In this case, a nearby point is returned as the return result by using the one-sided number.

[0176] 1) Check if the main point of the text to be searched is a road type. If yes, continue, otherwise end;

[0177] 2) Determine the parity of the sub-point value of the text to be searched;

[0178] 3) According to the parity of the sub-point value of the text to be searched, filter out the sub-points in the second recall text list that are inconsistent with the parity of the sub-point of the text to be searched;

[0179] 4) Filter the second recall text sub-point and the text to be searched sub-point number difference. If there is no nearby point at this time, end, otherwise continue;

[0180] 5) Select the second recall text sub-point number closest to the sub-point number of the text to be searched, and find the second recall text data corresponding to the number;

[0181] 6) If the data corresponding to the left number (or the right number) is greater than or equal to 3, the abnormal point needs to be removed. The specific method is to calculate the center point according to the mean of the latitude and longitude of the second recall text data, and filter out the point with large distance from the center.

[0182] 7) The left number (or the right number) adopts the filtered point position to calculate the average value, and the final latitude and longitude coordinates corresponding to the number are obtained;

[0183] The confidence is calculated, the similarity and the number offset are weighted and summed to obtain the confidence, and the specific formula is as follows

[0184] offset = |num near -num pred |-1 (5)

[0185] confidence = similarity + (1-2*offset / threshold)*0.05 (6)

[0186] Wherein, num represents the number, offset represents the absolute value of the difference between the two point numbers and the difference between the offset; threshold is the number threshold; the confidence formula is the sum of the text similarity and the number difference weight.

[0187] Example:

[0188] The text number to be searched is 14, and the second recall text number list is [11, 13, 19]

[0189] 13 can be found in the second recall text, so the latitude and longitude corresponding to the number 13 are calculated by using the above strategy, and the latitude and longitude and data are returned as the return result.

[0190] Method four:

[0191] The cell double-sided number generation strategy refers to when the main point component in the text to be searched is POI (level 6 and level 10) and the predicted sub-point number component label is (level 26 and level 22), the second recall text can find the numbers adjacent to the sub-point number of the text to be searched. In this case, a virtual point is generated by the two numbers as the return result.

[0192] 1) Filter the second recall text sub-point and the text to be searched sub-point number difference. If there is only a single adjacent point at this time, end, otherwise continue;

[0193] 2) Select the two second recall text sub-point numbers closest to the text to be searched sub-point number, and find the second recall text data corresponding to the number;

[0194] 3) If the data corresponding to the left number (or the right number) is greater than or equal to 3, the abnormal point needs to be removed. The specific method is to calculate the center point of the second recall text data latitude and longitude, and filter the point position with large distance difference from the center;

[0195] 4) Left number (or right number) uses the filtered point position to calculate the average value, and the final latitude and longitude coordinates corresponding to the number are obtained;

[0196] Based on the final latitude and longitude coordinates of the left and right adjacent numbers, the latitude and longitude of the middle number (the text sub-point number to be searched) are predicted by linear interpolation. The specific formula is as follows:

[0197]

[0198]

[0199] The confidence is calculated, and the weighted sum of the similarity and the distance is obtained. The specific formula is as follows

[0200] offset = |num near -num pred |-1 (9)

[0201] confidence = similarity + (1-2*offset / threshold)*0.05 (10)

[0202] Method five:

[0203] The cell single number generation strategy refers to that when the main point component in the text to be searched is a POI (level 6 and level 10) and the predicted sub-point number component label is (level 26 and level 22), the single adjacent number of the sub-point number of the text to be searched can be found in the second recall text. In this case, a single adjacent point is returned as the return result.

[0204] 1) Filter the second recall text sub-point and the text to be searched sub-point number difference. If there is no adjacent point at this time, end, otherwise continue;

[0205] 2) Select the second recall text sub-point number closest to the text to be searched sub-point number, and find the second recall text data corresponding to the number;

[0206] 3) If the data corresponding to the left number (or right number) is greater than or equal to 3, the abnormal point needs to be removed. The specific method is to calculate the center point of the second recall text data latitude and longitude according to the average value, and filter the point position with large distance difference from the center;

[0207] 4) Left number (or right number) uses the filtered point position to calculate the average value, and the final latitude and longitude coordinates corresponding to the number are obtained;

[0208] The confidence is calculated, and the weighted sum of the similarity and the number offset is obtained. The specific formula is as follows:

[0209] offset = |numnear num pred |-1 (11)

[0210] confidence = similarity + (1 - 2 * offset / threshold) * 0.05 (12)

[0211] It should be noted that when the output result of the adjacent point is an empty set, the first target entity in the text to be searched is retained, and the target place corresponding to the text to be searched is output according to the first target entity. Specifically, the main point in the text to be searched can be split, the component filtering is performed according to the main point component, the subsequent sub-point part is removed, then the third recall text is filtered, the original data set is the third recall text of the main point, the third recall text is filtered according to the pre-processed main sub-point type, the third recall text of the type of the main point is retained, finally, the main point prediction is performed based on the third recall text. The purpose of this scheme is to provide a backup when the sub-point prediction returns no result. The specific method can be referred to the process of the main point prediction described above, and will not be repeated here.

[0212] It should be further noted that different search scenarios require different place accuracy. For example, for a taxi or freight scenario, it only needs to be accurate to the road or community place; and for a food delivery scenario, it needs to be accurate to a specific house number. Therefore, in some embodiments, before the step of "outputting the target place corresponding to the text to be searched based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier", specifically, it can include:

[0213] (61) obtaining a current place search environment;

[0214] (62) retaining the third target entity in the text to be searched based on the current place search strategy.

[0215] Optionally, in some embodiments, the step of "outputting the target place corresponding to the text to be searched based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier" can be specifically: outputting the target place corresponding to the text to be searched based on the entity identifier corresponding to the third target entity and the identifier type corresponding to the entity identifier.

[0216] The process of searching for a target place will be described in the previous embodiments, and will not be repeated here.

[0217] After obtaining the text to be searched, the embodiment of the application performs entity segmentation on the text to be searched to obtain at least one entity corresponding to the text to be searched. Then, according to a preset association relationship, the embodiment of the application determines the entity identifiers corresponding to each entity in the text to be searched. Finally, based on the entity identifiers corresponding to each entity and the identifier types corresponding to the entity identifiers, the embodiment of the application outputs a target location corresponding to the text to be searched. The scheme for location searching provided by the application can perform entity segmentation on the text to be searched to reduce the influence of meaningless words on the search results. Moreover, based on the entity identifiers corresponding to each entity and the identifier types corresponding to the entity identifiers, the embodiment of the application outputs a target location corresponding to the text to be searched. Therefore, the scheme for location searching provided by the application can not only improve the locations recalled during location searching, but also improve the accuracy of location searching.

[0218] To better implement the location searching method of the embodiment of the application, the embodiment of the application further provides a location searching device based on the above-mentioned location searching device (referred to as a searching device). The meanings of the terms are the same as those in the above-mentioned location searching method, and specific implementation details can be referred to the description in the method embodiment.

[0219] Please refer to Figure 3 , Figure 3 The structure diagram of the location searching device provided by the embodiment of the application is shown in FIG. 1. The training device can include an acquisition module 201, a segmentation module 202, a determination module 203, and an output module 204, and the specific implementation details can be as follows.

[0220] The acquisition module 201 is configured to acquire the text to be searched.

[0221] The text to be searched can be a text input by a user, a text saved in a local database, or a text obtained by accessing a network interface. The specific implementation details depend on the actual situation.

[0222] The segmentation module 202 is configured to perform entity segmentation on the text to be searched to obtain at least one entity corresponding to the text to be searched.

[0223] For example, specifically, a preset named entity recognition (NER) algorithm can be used to perform entity segmentation on the text to be searched to obtain at least one entity corresponding to the text to be searched. NER is a very basic task in natural language processing, and is an important basic tool for numerous NLP tasks such as information extraction, question and answer systems, syntax analysis, and machine translation. Named entities generally refer to entities in a text that have specific meanings or strong referentiality. In academia, they usually include three categories of entity classes, time classes, and digital classes, and seven subcategories of personal names, place names, organization names, time, date, currency, and percentage. NER is to extract the above-mentioned entities from unstructured input text, and can identify more categories of entities according to business needs.

[0224] The determining module 203 is configured to determine the entity identifier corresponding to each entity in the text to be searched according to the preset association relationship.

[0225] After the entity segmentation of the text to be searched, the determining module 203 can determine a corresponding entity identifier for each entity according to the preset association relationship.

[0226] The output module 204 is configured to output the target location corresponding to the text to be searched based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier.

[0227] For example, specifically, for the text to be searched "Beijing Changping District, Innovation Road No. 11, 10th Floor, Room 1015", it is known from Table 1 that the corresponding identifier is "2.4.8.21.23.24.25", and thus the target location corresponding to the text to be searched can be obtained.

[0228] Optionally, in some embodiments, the output module 204 can specifically include:

[0229] The identifying unit is configured to identify the identifier type corresponding to the entity identifier.

[0230] The determining unit is configured to determine the location search strategy corresponding to the text to be searched based on the identifier type.

[0231] The output unit is configured to output the target location corresponding to the text to be searched according to the determined location search strategy and the entity identifier corresponding to each entity.

[0232] Optionally, in some embodiments, the output unit can specifically include:

[0233] The obtaining sub-unit is configured to obtain the administrative region corresponding to the text to be searched.

[0234] The first determining sub-unit is configured to determine a first target entity among the entities based on the first location search strategy.

[0235] The first output sub-unit is configured to output the target location corresponding to the text to be searched based on the administrative region and the entity identifier corresponding to the first target entity.

[0236] Optionally, in some embodiments, the output sub-unit can be specifically configured to determine a first recall text corresponding to the text to be searched in a preset text library based on the administrative region and the entity identifier corresponding to the first target entity, determine a first target text that meets a preset condition as the first target text, and determine the location corresponding to the first target text as the target location corresponding to the text to be searched.

[0237] Optionally, in some embodiments, the output unit can specifically include:

[0238] Construct sub-units to build second recall texts corresponding to the text to be retrieved based on the second location retrieval strategy;

[0239] The calculation subunit is used to calculate the edit distance similarity between the text to be retrieved and each of the second recalled texts;

[0240] The second determining subunit is used to determine the second recalled text with an edit distance similarity greater than a preset value as candidate text;

[0241] The output sub-unit is used to determine the candidate text with the highest edit distance similarity as the second target text, and to determine the location corresponding to the second target text as the target location corresponding to the text to be retrieved.

[0242] Optionally, in some embodiments, the output module may also be used to: predict the nearest locations of each entity when the prediction result of the second location retrieval strategy is an empty set; and output the target location corresponding to the text to be retrieved based on the prediction result.

[0243] Optionally, in some embodiments, the output module may also be used to: when the output result of the neighboring points is an empty set, retain the first target entity in the text to be retrieved, and output the target location corresponding to the text to be retrieved based on the first target entity.

[0244] In this embodiment, after the acquisition module 201 acquires the text to be retrieved, the segmentation module 202 performs entity segmentation on the text to obtain at least one entity corresponding to the text. Then, the determination module 203 determines the entity identifiers corresponding to each entity in the text according to a preset association relationship. Finally, the output module 204 outputs the target location corresponding to the text based on the entity identifiers and identifier types corresponding to each entity. The location retrieval scheme provided in this application can segment the text to be retrieved to reduce the impact of meaningless words on the retrieval results. Furthermore, by outputting the target location corresponding to the text based on the entity identifiers and identifier types corresponding to each entity, it can not only improve the location recall during location retrieval but also improve the accuracy of location retrieval.

[0245] Furthermore, embodiments of this application also provide an electronic device, such as... Figure 4 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically:

[0246] The electronic device may include components such as a processor 301 with one or more processing cores, a memory 302 with one or more computer-readable storage media, a power supply 303, and an input unit 304. Those skilled in the art will understand that... Figure 4The electronic device structure shown in the figure does not constitute a limitation on the electronic device, and can include more or fewer components than shown, or combine certain components, or arrange different components. Among them:

[0247] The processor 301 is the control center of the electronic device, connects various parts of the entire electronic device through various interfaces and lines, and performs various functions of the electronic device and processes data by running or executing software programs and / or modules stored in the memory 302, and calling data stored in the memory 302, thereby overall monitoring the electronic device. Optionally, the processor 301 can include one or more processing cores; preferably, the processor 301 can integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application program, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 301.

[0248] The memory 302 can be used to store software programs and modules, and the processor 301 executes various functions and applications and location retrieval by running the software programs and modules stored in the memory 302. The memory 302 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; the data storage area can store data created according to the use of the electronic device, etc. In addition, the memory 302 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory device. Accordingly, the memory 302 can also include a memory controller to provide access for the processor 301 to the memory 302.

[0249] The electronic device also includes a power supply 303 for supplying power to various components, and preferably the power supply 303 can be logically connected to the processor 301 through a power management system, so as to realize functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 303 can also include one or more than one direct current or alternating current power supply, a recharging system, a power supply failure detection circuit, a power supply converter or inverter, a power supply state indicator, and any other components.

[0250] The electronic device can also include an input unit 304, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0251] Although not shown, the electronic device can further include a display unit and the like, which will not be described herein. Specifically, in the present embodiment, the processor 301 in the electronic device will load the executable file corresponding to the process of one or more application programs into the memory 302 according to the following instructions, and run the application program stored in the memory 302 by the processor 301, thereby realizing various functions, as follows:

[0252] Obtaining a text to be searched; performing entity segmentation on the text to be searched to obtain at least one entity corresponding to the text to be searched; determining entity identifiers corresponding to each entity in the text to be searched according to a preset association relationship; and outputting a target location corresponding to the text to be searched based on the entity identifiers corresponding to each entity and the identifier types corresponding to the entity identifiers.

[0253] The specific implementation of each operation can be referred to the foregoing embodiments, which will not be described herein.

[0254] After obtaining the text to be searched, the embodiment of the present application performs entity segmentation on the text to be searched to obtain at least one entity corresponding to the text to be searched, then determines entity identifiers corresponding to each entity in the text to be searched according to a preset association relationship, and finally outputs a target location corresponding to the text to be searched based on the entity identifiers corresponding to each entity and the identifier types corresponding to the entity identifiers. The scheme of the present application for location searching can perform entity segmentation on the text to be searched to reduce the influence of meaningless words on the search results, and output a target location corresponding to the text to be searched based on the entity identifiers corresponding to each entity and the identifier types corresponding to the entity identifiers, thereby not only improving the recalled locations during location searching, but also improving the accuracy of location searching.

[0255] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or controlled by instructions related to hardware, which can be stored in a computer readable storage medium and loaded and executed by a processor.

[0256] To this end, the embodiment of the present application provides a storage medium having a plurality of instructions stored therein, which can be loaded by a processor to execute the steps in any one of the location searching methods provided by the embodiments of the present application. For example, the instructions can execute the following steps:

[0257] Obtaining a text to be searched; performing entity segmentation on the text to be searched to obtain at least one entity corresponding to the text to be searched; determining entity identifiers corresponding to each entity in the text to be searched according to a preset association relationship; and outputting a target location corresponding to the text to be searched based on the entity identifiers corresponding to each entity and the identifier types corresponding to the entity identifiers.

[0258] The specific implementation of the above operations can refer to the foregoing embodiments, which will not be described herein again.

[0259] The storage medium can include a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, etc.

[0260] Due to the instructions stored in the storage medium, the steps of any of the location retrieval methods provided in the embodiments of the present application can be executed, thus achieving the beneficial effects of any of the location retrieval methods provided in the embodiments of the present application. Details are described in the foregoing embodiments, which will not be described herein again.

[0261] The foregoing describes in detail the location retrieval method, device, electronic equipment and storage medium provided in the embodiments of the present application. The principles and implementation manners of the present application are described by applying specific examples in this paper. The foregoing embodiment description is only used to help understand the method and core idea of the present application. Meanwhile, for those skilled in the art, the specific implementation manners and application ranges will be changed according to the idea of the present application. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A place search method characterized by comprising: The method comprises the following steps: acquiring a text to be searched; performing entity segmentation on the text to be searched to obtain at least one entity corresponding to the text to be searched, the entity being a word having a specific semantic meaning; determining entity identifiers corresponding to each entity in the text to be searched according to a preset association relationship, the association relationship being an association relationship between levels, place names and commonly used keywords; outputting a target place corresponding to the text to be searched based on the entity identifiers corresponding to each entity and the identifier types corresponding to the entity identifiers, specifically comprising: identifying the identifier types corresponding to the entity identifiers; determining a place searching strategy corresponding to the text to be searched based on the identifier types; outputting a target place corresponding to the text to be searched according to the determined place searching strategy and the entity identifiers corresponding to each entity; the step of outputting a target place corresponding to the text to be searched according to the determined place searching strategy and the entity identifiers corresponding to each entity comprises: constructing a second recall text corresponding to the text to be searched based on a second place searching strategy; calculating an edit distance similarity between the text to be searched and each second recall text; determining a second recall text with an edit distance similarity greater than a preset value as a candidate text; determining a candidate text with the greatest edit distance similarity as a second target text, and determining a place corresponding to the second target text as a target place corresponding to the text to be searched; when the prediction result of the second place searching strategy is an empty set, performing a nearby point prediction on each entity; outputting a target place corresponding to the text to be searched based on the prediction result.

2. The method of claim 1, wherein, the step of outputting a target place corresponding to the text to be searched according to the determined place searching strategy and the entity identifiers corresponding to each entity comprises: acquiring an administrative region corresponding to the text to be searched; determining a first target entity among each entity based on a first place searching strategy; outputting a target place corresponding to the text to be searched based on the administrative region and the entity identifier corresponding to the first target entity.

3. The method of claim 2, wherein, the step of outputting a target place corresponding to the text to be searched based on the administrative region and the entity identifier corresponding to the first target entity comprises: determining a first recall text corresponding to the text to be searched in a preset text library based on the administrative region and the entity identifier corresponding to the first target entity; determining a first target text meeting a preset condition as the first target text, and determining a place corresponding to the first target text as a target place corresponding to the text to be searched.

4. The method of claim 1, wherein, further comprising: when the prediction result of the nearby point is an empty set, retaining the first target entity in the text to be searched; outputting a target place corresponding to the text to be searched according to the first target entity.

5. The method according to any one of claims 1 to 4, characterized in that, before the step of outputting a target place corresponding to the text to be searched based on the entity identifiers corresponding to each entity and the identifier types corresponding to the entity identifiers, further comprising: acquiring a current place searching environment; retaining a third target entity in the text to be searched based on the current place searching strategy; The outputting of the target place corresponding to the text to be searched based on the entity identifier corresponding to each entity and the identifier type corresponding to the entity identifier comprises: outputting the target place corresponding to the text to be searched based on the entity identifier corresponding to the third target entity and the identifier type corresponding to the entity identifier.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, The processor implements the steps of the place searching method according to any one of claims 1-5 when executing the program.

7. A computer-readable storage medium, characterized in that, A computer program is stored thereon, and the computer program implements the steps of the place searching method according to any one of claims 1-5 when executed by a processor.

Images