Information Push Method, Device and Server

Abstract

The present invention provides an information push method, device, and server, wherein the method comprises: obtaining information to be pushed; extracting events from the information to be pushed to obtain key events and first target sentences related to the key events in the information to be pushed; screening the key events based on the user's historical behavior data to obtain target events; generating target information based on the first target sentence corresponding to the target event; and pushing the target information to the user. Because the generated target information is determined based on the user's historical behavior data, different users can receive different target information for the same information to be pushed, thereby meeting the user's personalized needs, making the target information more in line with the user's needs, and improving the push effect.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of information push technology, and in particular to an information push method, device and server. Background Art

[0002] With the advancement of society and technology, people's pace of life is accelerating. At the same time, people's consumption, entertainment, daily life, and learning are becoming increasingly personalized. Universal content or methods are difficult to meet everyone's needs and preferences. This is especially true in today's era of rapid information development, where all types of information are growing exponentially. Because everyone has different knowledge backgrounds, behavioral habits, and interests, their interest in content and how they read it also differ.

[0003] Current news and information products generally combine user behavior preferences to make personalized recommendations at the information article level. This meets the user's personalized needs to a certain extent, but for the same information article, the content pushed is the same, and the push effect is poor. Summary of the Invention

[0004] The embodiments of the present invention provide an information push method, device and server to solve the problem in the prior art that the information pushed is identical, resulting in poor push effect.

[0005] To solve the above problems, the present invention is achieved as follows:

[0006] In a first aspect, an embodiment of the present invention provides an information push method, the method comprising:

[0007] Get the selected push information;

[0008] Performing event extraction on the to-be-selected push information to obtain a key event and a first target sentence related to the key event in the to-be-selected push information;

[0009] Filter the key events based on the user's historical behavior data to obtain target events;

[0010] generating target information according to a first target sentence corresponding to the target event;

[0011] Push the target information to the user.

[0012] In a second aspect, an embodiment of the present invention further provides an information push device, comprising:

[0013] The acquisition module is used to obtain the push information to be selected;

[0014] An event extraction module, configured to extract events from the selected push information to obtain a key event and a first target sentence related to the key event in the selected push information;

[0015] A screening module is used to screen the key events based on the user's historical behavior data to obtain target events;

[0016] A generating module, configured to generate target information according to a first target sentence corresponding to the target event;

[0017] A push module is used to push the target information to the user.

[0018] In a third aspect, an embodiment of the present invention further provides a server comprising: a transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; the processor is configured to read the program in the memory to implement the steps of the method described in the first aspect above.

[0019] In a fourth aspect, an embodiment of the present invention further provides a readable storage medium for storing a program, which, when executed by a processor, implements the steps in the method described in the first aspect.

[0020] In an embodiment of the present invention, a method is used to obtain information to be pushed; extract events from the information to be pushed to obtain key events and a first target sentence related to the key events in the information to be pushed; filter the key events based on the user's historical behavior data to obtain target events; generate target information based on the first target sentence corresponding to the target event; and push the target information to the user. Because the generated target information is determined based on the user's historical behavior data, different users will receive different target information for the same information to be pushed, which can meet the user's personalized needs, making the target information more in line with the user's needs and improving the push effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in describing the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0022] Figure 1 This is one of the flow charts of the information push method provided by an embodiment of the present invention;

[0023] Figure 2 This is a schematic structural diagram of an information push device provided by an embodiment of the present invention;

[0024] Figure 3 This is the second flow chart of the information push method provided by the embodiment of the present invention;

[0025] Figure 4 It is another structural diagram of the information push device provided by the embodiment of the present invention;

[0026] Figure 5 It is a structural diagram of a server provided by the implementation of the present invention. DETAILED DESCRIPTION

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0028] The terms "first", "second" etc. in the embodiments of the present invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or equipment comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or equipment. In addition, "and / or" is used in this application to represent at least one of the connected objects, for example A and / or B and / or C, which means comprising 7 situations including single A, single B, single C, and both A and B exist, both B and C exist, both A and C exist, and both A, B and C exist.

[0029] See also Figure 1 , Figure 1 This is one of the flow charts of the information push method provided by an embodiment of the present invention. Figure 1 The information push method shown may be executed by a server, for example, a server for information push, and may specifically include the following steps:

[0030] Step 101: Obtain push information to be selected.

[0031] The selected push information may be determined based on the user's historical behavior data. For example, if the user's historical behavior data indicates that the user frequently reads football and entertainment news information, football and entertainment news information may be selected as the selected push information. The selected push information may be text information, including multiple sentences.

[0032] Specifically, a user's historical behavior data may include information reading behavior, information content data, and other data (such as basic user information). Based on this multi-dimensional data, the user's information interest preferences can be generated. Based on this user's preferences, information articles that match the user's preferences are found in the information library and a list of recommended information is generated. Each recommended list corresponds to a candidate for push notification.

[0033] Step 102: extract events from the selected push information to obtain key events and a first target sentence related to the key events in the selected push information.

[0034] Event extraction is performed on the selected push information to obtain key events. For example, event extraction can be performed based on a real event extraction algorithm. Specifically, event extraction can be performed based on annotated corpus, such as training a machine learning model or a deep learning model based on the annotated corpus, and using the trained machine learning model or deep learning model to perform event extraction on the selected push information; or a rule-based template extraction method can be used for event extraction, such as event extraction based on text semantic rules, text description rules, etc. The specific event extraction method can be flexibly selected according to actual conditions and is not limited here.

[0035] By performing event extraction on the selected push information, one or more key events can be obtained. Based on each key event, a first target sentence related to the key event is obtained from the selected push information. That is, the first target sentence is a sentence in the selected push information that is associated with the key event. For example, the first target sentence may reflect the information expressed in the key event. The first target sentence may include one or more sentences.

[0036] Step 103: Filter the key events according to the user's historical behavior data to obtain target events.

[0037] A user's historical behavior data may include the user's information reading behavior data, information content data, and other data (such as user basic information, etc.). Based on the historical user behavior data, key events are screened to obtain target events. For example, based on the user's historical behavior data, it is determined that the user frequently reads financial information. Based on the selected push information, the key events determined include the author's resume introduction event and the financial market introduction event. When screening key events, the financial market introduction event can be used as the target event. The target event can be one key event or multiple key events.

[0038] Step 104: Generate target information according to the first target sentence corresponding to the target event.

[0039] After determining the target event, the first target sentence corresponding to the target event can be obtained, and the target information can be generated based on the first target sentence. For example, the first target sentence can be directly used as the target information, or the target information can be obtained after deduplication processing of the text generated by the first target sentence, or the title of the to-be-selected push information or the picture in the to-be-selected push information can be added to the text generated by the first target sentence after deduplication processing to obtain the target information.

[0040] Step 105: Push the target information to the user.

[0041] The user in this application can be understood as a user who uses a terminal. Pushing target information to the user means sending the target information to the terminal. The target information can be displayed on the terminal for the user to read. The terminal can be a mobile phone, tablet computer, laptop and other terminal devices.

[0042] In this embodiment, the information to be pushed is obtained; events are extracted from the information to be pushed to obtain key events and the first target sentence related to the key events in the information to be pushed; the key events are screened based on the user's historical behavior data to obtain target events; target information is generated based on the first target sentence corresponding to the target event; and the target information is pushed to the user. In the above, by performing event extraction on the information to be pushed, and screening the extracted key events based on the user's historical behavior data, the first target sentence corresponding to the target event is obtained, and target information is generated based on the first target sentence. The target information generated in this way is determined based on the user's historical behavior data. For the same information to be pushed, the target information pushed to different users is different, which can meet the personalized needs of users, make the target information more in line with the needs of users, and improve the push effect.

[0043] In one embodiment of the present application, after step 101, obtaining the selected push information, and before step 104, generating the target information according to the first target sentence corresponding to the target event, the process further includes:

[0044] Obtaining the target importance of each sentence in the selected push information;

[0045] Determining a second target sentence based on the target importance of each sentence in the selected push information, where the second target sentence is the top N target importance sentences sorted from high to low in order of importance, where N is a positive integer;

[0046] Accordingly, step 104, generating target information according to the first target sentence corresponding to the target event, includes:

[0047] Target information is generated according to the second target sentence and the first target sentence corresponding to the target event.

[0048] In this embodiment, the target information is generated not only based on the first target sentence corresponding to the target event, but also based on the target importance of each sentence in the selected push information. The process of determining the target importance of a sentence is described below.

[0049] In the above, the target importance of a sentence can be understood as the importance of the sentence in the selected push information. Specifically, it can be determined based on the semantic similarity between the sentence and other sentences, or based on the semantic similarity between the sentence and the title of the selected push information, or based on the similarity between the sentence and the user's keywords. Specifically, calculating the target importance of each sentence in the selected push information includes:

[0050] Calculating the first importance of each sentence in the selected push information based on the semantic similarity between the sentences;

[0051] determining the second importance of each sentence according to the similarity between each sentence and the title of the selected push information;

[0052] determining a third importance of each sentence based on a similarity between each sentence and a keyword of the user, wherein the keyword of the user is a keyword of a browsing content preference determined based on the historical behavior data;

[0053] A target importance of each sentence is determined based on the first importance, the second importance, and the third importance of each sentence.

[0054] In the above, the sentence vector of each sentence in the selected push information is first obtained, and then the semantic similarity between any two sentences is calculated, specifically using cosine similarity. A graph is established based on sentence similarity, where the vertices in the graph represent sentences and the edges in the graph represent the similarity between sentences. Then, the importance score of each sentence, i.e., the first importance, is calculated based on the semantic similarity between sentences, including:

[0055] Splitting the to-be-selected push information to obtain multiple words;

[0056] Obtain word vectors for the multiple words;

[0057] Obtaining a sentence vector of a third target sentence according to the word vectors of each word in the third target sentence, wherein the third target sentence is any one of the sentences in the to-be-selected push information;

[0058] determining, based on the sentence vectors of any two third target sentences in the to-be-selected push information, the semantic similarity of the any two third target sentences;

[0059] The first importance of each third target sentence is determined according to the semantic similarity between any two third target sentences in the to-be-selected push information.

[0060] In the above, the selected push information is split to obtain multiple words. For each word, the word vector (word2vec) model or BERT model can be used to obtain the word vector of each word. For example, if the word vector of the i-th word in a sentence is represented by [w i1 ,w i2 ,......,w in ], then based on the word vector, generate the vector representation of each sentence, that is, the sentence vector. Sentence vector generation can be done by taking the average value of each word vector in the sentence. If a sentence vector is represented as [S1, S2, ..., S n ],in:

[0061]

[0062] In the above, m is the number of word vectors included in the sentence, and n is the vector dimension.

[0063] The sentence vectors of any two third target sentences in the selected push information are obtained through the above method. Based on the sentence vectors, the semantic similarity between any two third target sentences is calculated. Specifically, cosine similarity can be used for calculation. A graph is established based on the sentence similarity, where the vertices in the graph represent the third target sentences and the edges in the graph represent the similarity between the third target sentences. Then, the importance score of each third target sentence is calculated based on the TextRank algorithm, that is, the first importance. The calculation expression is as follows:

[0064]

[0065] Among them, w jk It represents the similarity between sentence j and sentence k, d is the damping coefficient, which ranges from 0 to 1, and TR is the first importance of the third target sentence. The above formula is iterated and finally a stable TR value can be obtained. n is the dimension of the sentence vector, In(v i ) represents the sentence similarity-based graph with v i The set of vertices connected by edges, Out(v j ) represents the sentence similarity-based graph with v j A set of vertices connected by edges.

[0066] When calculating the second importance of each sentence, each sentence and the title of the selected push information can be calculated using the above-mentioned method of calculating the first importance, that is, first calculate the semantic similarity between each sentence and the title, and then determine the second importance of each sentence based on the similarity.

[0067] Similarly, when calculating the third importance of each sentence, each sentence and the user's keyword can be calculated using the same method as for calculating the first importance described above. That is, the semantic similarity between each sentence and the keyword is first calculated, and then the third importance of each sentence is determined based on the similarity. The user's keyword is a keyword associated with the browsing content preference determined based on the historical behavior data. For example, the user's information interest preferences are generated based on historical behavior data such as the user's information reading behavior and information content, and keywords are determined based on these interest preferences.

[0068] The first importance and the second importance can represent the importance of the sentence in the selected push information, and the third importance can represent the degree of correlation between the sentence and the user's interest preferences. The target importance obtained by the first importance, the second importance and the third importance can comprehensively consider the importance of the sentence in the selected push information and the degree of correlation between the sentence and the user's interest preferences. The target importance can more accurately reflect whether the sentence meets the user's needs.

[0069] When determining the target importance of each sentence based on the first importance, the second importance and the third importance of each sentence, the first importance, the second importance and the third importance can be averaged, or weights can be set for the first importance, the second importance and the third importance respectively, and the target importance can be obtained by taking a weighted average of the first importance, the second importance and the third importance. The specific selection can be made according to actual conditions and is not limited here.

[0070] Furthermore, in one embodiment of the present application, generating target information according to the second target sentence and the first target sentence corresponding to the target event includes:

[0071] generating an initial text based on the second target sentence and the first target sentence corresponding to the target event;

[0072] Calculating a target similarity between a first sentence and a second sentence in the initial text, where the first sentence and the second sentence are any two sentences in the initial text;

[0073] If the target similarity is greater than a first preset threshold, removing the first sentence or the second sentence from the initial text to obtain a target text;

[0074] The target information is obtained according to the target text.

[0075] After determining the second target sentence based on the target importance of the sentences in the to-be-selected push information, and performing event extraction on the to-be-selected push information to obtain the first target sentence corresponding to the target event, an initial text is generated based on the first target sentence and the second target sentence. For example, the order of each target sentence in the initial text can be determined according to the position order of each target sentence in the to-be-selected push information.

[0076] In order to further refine the initial text and improve the quality of the initial text, the initial text can be subjected to sentence deduplication processing, that is, the target similarity between the first sentence and the second sentence in the initial text is calculated. If the target similarity is greater than a first preset threshold, it means that the first sentence and the second sentence are highly similar and may be repeated sentences. The first sentence or the second sentence is then removed from the initial text to obtain the target text. That is, only the first sentence or the second sentence is retained in the target text to avoid sentence redundancy and affect the quality of the target information.

[0077] Furthermore, target information is obtained based on the target text. For example, the title of the selected push information or the picture in the selected push information is added to the target text to obtain the target information.

[0078] Furthermore, in order to improve the quality of the target information finally obtained, not only can the initial text be subjected to sentence deduplication processing, but also sentence error correction processing can be performed. That is, if the target similarity is greater than a first preset threshold, the first sentence or the second sentence is removed from the initial text to obtain the target text, including:

[0079] If the target similarity is greater than a first preset threshold, removing the first sentence or the second sentence from the initial text to obtain an intermediate text;

[0080] Calculating a probability of a third sentence in the intermediate text, where the third sentence is any sentence in the intermediate text;

[0081] If the probability is less than a second preset threshold, error correction is performed on the third sentence in the intermediate text to obtain the target text.

[0082] For example, the N-gram model can be used to calculate the probability of the third sentence, and error correction processing can be performed on sentences with lower probabilities to improve the accuracy of sentences in the target text and improve the quality of the target text.

[0083] Furthermore, in order to improve the hit rate of target information and enhance the recommendation effect, the historical behavior data may be updated based on the user's behavior data for the target information. That is, after pushing the target information to the user, the method further includes:

[0084] Recording the current behavior data of the user operating the target information;

[0085] The historical behavior data is updated using the current behavior data.

[0086] Historical behavior data may include current behavior data for the target information, such as whether it has been browsed, browsing duration, number of views, whether the target information has been forwarded, etc. This helps the server determine whether the target information push is successful based on the current behavior data, and adjust the information pushed next time based on the current behavior data, continuously optimize the pushed information, and improve the push accuracy.

[0087] For easier understanding, the following examples are provided:

[0088] Figure 2 The figure shows a module diagram of the information push device provided by this application, which includes the following modules:

[0089] Data collection module: including user information reading behavior data, information content data, and other data (such as user basic information, etc.);

[0090] User Preference and Recommendation Calculation Module: This module generates user information interest preferences based on multi-dimensional data such as user information reading behavior and information content data. Based on user preferences, it finds information articles that match the user's preferences from the information library and generates an information recommendation list.

[0091] Personalized content generation module: Generates personalized news content for users based on their interests and the content of listed news articles. This module includes generating news summaries tailored to user interests, generating key event content tailored to user concerns, and smoothing news content.

[0092] Application module: Personalized information content push and display module: Push personalized information content generated for users to users and display it.

[0093] Among them, the data acquisition module includes the following types of data:

[0094] User behavior data, i.e., information reading behavior: including clicks, reading time, location of articles read, number of visits, etc.;

[0095] Information content data: information classification, labels, content, etc.;

[0096] Other user data: such as basic user information, including age, region, gender and other basic information data. This data is optional.

[0097] The user preference and recommendation calculation module includes: a user interest preference calculation module and an information recommendation list generation module.

[0098] The user interest preference calculation module is used to generate the user's information interest preferences based on the user's information reading behavior, information content, other information and other data.

[0099] The information recommendation list generation module is used to find information articles that meet user preferences from the information library based on user preferences and generate a list of recommended articles for the user.

[0100] The personalized information content generation module is used to generate personalized information content for each information article in the recommendation list based on the user's interest preferences. The overall process is as follows Figure 3 shown.

[0101] The user-interest-based news summary generation module calculates the importance of sentences in news articles by comprehensively considering factors such as sentence similarity, similarity with the title, and position in the article. It also integrates user interests to comprehensively determine sentence importance, extract the K most important sentences, and generate news summary content that incorporates user interests. This module specifically includes the following steps:

[0102] (1) Calculate sentence importance based on sentence semantic similarity.

[0103] Step A: Get the text contained in the news article. Split the text into individual sentences and words;

[0104] Step B: Use word2vec model or bert model to obtain the word vector of each word in the text. If the word vector of the i-th word in a sentence is represented by [w i1 ,w i2 ,......,w in ].

[0105] Step C: Generate a vector representation of each sentence based on the word vector, i.e., the sentence vector. Sentence vector generation can be achieved by averaging the vectors of each word in the sentence. Suppose a sentence vector is represented as [S1, S2, ..., S n ],in:

[0106]

[0107] Step D: Based on the sentence vectors, calculate the similarity between the two sentences, and use a similarity calculation method such as cosine similarity;

[0108] Step E: Create a graph based on sentence similarity, where vertices represent sentences and edges represent similarities between sentences. Calculate sentence importance scores based on TexRank for subsequent sentence importance ranking. The formula is as follows:

[0109]

[0110] Among them, w jk It represents the similarity between sentence j and sentence k, d is the damping coefficient, which ranges from 0 to 1, and TR is the first importance of the third target sentence. The above formula is iterated and finally a stable TR value can be obtained. n is the dimension of the sentence vector, In(v i ) represents the sentence similarity-based graph with v i The set of vertices connected by edges, Out(v j ) represents the sentence similarity-based graph with v j A set of vertices connected by edges.

[0111] (2) Calculate sentence importance by adding factors such as the similarity between the sentence and the title and the position. The similarity calculation method can refer to steps B to D above.

[0112] (3) Add the user content preference factor and calculate the similarity between the sentence and the user content preference keyword as an influencing factor for calculating the importance of the sentence.

[0113] (4) Based on the similarities calculated in the above three steps (1), (2) and (3), the final importance value of the sentence is calculated.

[0114] (5) Select K sentences and generate information summary content that is tailored to the user's interests.

[0115] The key event content generation module that meets the user's concerns is used to extract key events and the original text content descriptions corresponding to the events from the information content. Finally, based on the user's interests, the key events that the user is concerned about are selected to generate the information content for this part.

[0116] For event extraction algorithms, if there is no annotated corpus, rule-based template extraction methods such as text semantic rules and text description rules can be used; if there is annotated corpus, training methods based on machine learning and deep learning can also be used, such as dynamic multi-pooling convolutional neural network (DMCNN).

[0117] The following technical implementation uses the text semantic rule template method as an example. The main solution is: syntactic dependency analysis and semantic role labeling of information text, and event extraction based on the results of semantic role analysis and dependency syntactic analysis. The specific description is as follows:

[0118] Perform word segmentation, part-of-speech tagging, semantic role tagging, and dependency analysis on information text.

[0119] According to the above annotation results, trigger words are extracted to form a trigger word database, which provides a basis for subsequent event extraction.

[0120] Determine whether each sentence in the information content contains a trigger word. If so, extract event short sentences with semantic roles A0 and A1 based on the trigger word.

[0121] For the sentences without events extracted above, based on the results of dependency syntactic analysis, for each trigger word in the sentence, a short sentence with SBV and VOB dependency relationship is extracted.

[0122] If there is only a VOB relationship in the sentence and an ATT relationship modifies the verb, the word before the word is taken as the subject and extracted; if there are only SBV and CMP relationships in the sentence, the word modified by the CMP relationship (CMP verb-complement relationship) is extracted;

[0123] Extract the original text description corresponding to the event, usually using the context of the sentence where the event is located.

[0124] For the events obtained above, the similarity between the user's interests and the events is calculated, and the events with high similarity and the original content are output to generate key event content that meets the user's concerns.

[0125] The information content smoothing processing module is used to integrate the above-generated information summary content that is in line with user interests and the key event content that meets user concerns, and perform smoothing processing, including sentence deduplication, semantic coherence, and text error correction for similar content.

[0126] To remove duplicates from similar sentences, you can calculate the sentence similarity in steps B to D, filter out sentences with high similarity, remove redundant content, and generate the final personalized information content that meets the user's interests.

[0127] Semantic coherence and text error correction can be judged using the N-gram language model. Based on a massive text data corpus, the N-gram model is used to analyze the generation probability of information text and correct text with relatively low probability.

[0128] The personalized information content application module includes a personalized information content push and display module, which is used to push and display personalized information content generated for users. Based on user click behavior feedback, the effectiveness of personalized information content generation is evaluated and the personalized information content generation algorithm is continuously adjusted.

[0129] The above method generates user information content preferences based on the user's information reading behavior and information content, and generates an information recommendation list based on the user's preferences; secondly, for each information in the recommendation list, combined with the user's preferences, the summary information of the information content and the key event information that the user is concerned about are extracted to preliminarily form personalized information content; finally, after smoothing the information content, the final personalized information content is generated and pushed to the user for display and application. On the basis of pushing personalized information articles to users, it can further extract important information of the information content and the key event information that the user is concerned about based on the user's information reading behavior preferences, and generate personalized information content for users, thereby improving the efficiency of users in obtaining information and increasing the value of recommended information.

[0130] See also Figure 4 , Figure 4 This is a structural diagram of the information push device provided by an embodiment of the present invention. Figure 4 As shown, the information push device 400 includes:

[0131] The first acquisition module 401 is used to obtain the push information to be selected;

[0132] An event extraction module 402 is configured to extract events from the selected push information to obtain a key event and a first target sentence related to the key event in the selected push information;

[0133] The screening module 403 is used to screen the key events according to the user's historical behavior data to obtain target events;

[0134] A generating module 404 is configured to generate target information according to a first target sentence corresponding to the target event;

[0135] The push module 405 is configured to push the target information to the user.

[0136] Furthermore, the information push device 400 further includes:

[0137] A second acquisition module is used to obtain the target importance of each sentence in the selected push information;

[0138] a determination module, configured to determine a second target sentence based on the target importance of each sentence in the candidate push information, wherein the second target sentence is the sentences corresponding to the top N target importances sorted from high to low according to the target importance of each sentence in the candidate push information, where N is a positive integer;

[0139] The generating module 404 is configured to generate the target information according to the second target sentence and the first target sentence corresponding to the target event.

[0140] Furthermore, the second acquisition module includes:

[0141] A first determining submodule, configured to calculate a first importance of each sentence in the to-be-selected push information based on semantic similarity between the sentences;

[0142] A second determining submodule, configured to determine a second importance of each sentence based on a similarity between each sentence and the title of the selected push information;

[0143] a third determining submodule, configured to determine a third importance of each sentence based on a similarity between each sentence and a keyword of the user, wherein the keyword of the user is a keyword of a browsing content preference determined based on the historical behavior data;

[0144] The fourth determining submodule is configured to determine the target importance of each sentence according to the first importance, the second importance, and the third importance of each sentence.

[0145] Furthermore, the generating module 404 includes:

[0146] A first generating submodule, configured to generate an initial text based on the second target sentence and the first target sentence corresponding to the target event;

[0147] A calculation submodule, configured to calculate a target similarity between a first sentence and a second sentence in the initial text, where the first sentence and the second sentence are any two sentences in the initial text;

[0148] a deduplication submodule, configured to remove the first sentence or the second sentence from the initial text to obtain a target text if the target similarity is greater than a first preset threshold;

[0149] The acquisition submodule is used to obtain the target information according to the target text.

[0150] Furthermore, the deduplication submodule includes:

[0151] a first acquiring unit, configured to remove the first sentence or the second sentence from the initial text to obtain an intermediate text if the target similarity is greater than a first preset threshold;

[0152] a calculation unit, configured to calculate a probability of a third sentence in the intermediate text, where the third sentence is any sentence in the intermediate text;

[0153] An error correction unit is configured to correct the third sentence in the intermediate text to obtain the target text if the probability is less than a second preset threshold.

[0154] Furthermore, the first determining submodule includes:

[0155] a splitting unit, configured to split the to-be-selected push information to obtain a plurality of words;

[0156] A second acquiring unit, configured to acquire word vectors of the plurality of words;

[0157] a third acquiring unit, configured to acquire a sentence vector of a third target sentence based on word vectors of respective words in the third target sentence, wherein the third target sentence is any one of the sentences in the to-be-selected push information;

[0158] A first determining unit is configured to determine the semantic similarity of any two third target sentences in the to-be-selected push information based on the sentence vectors of the any two third target sentences;

[0159] The second determining unit is configured to determine the first importance of each third target sentence according to the semantic similarity between any two third target sentences in the to-be-selected push information.

[0160] Furthermore, the information push device 400 further includes:

[0161] Recording the current behavior data of the user operating the target information;

[0162] The historical behavior data is updated using the current behavior data.

[0163] The information push device 400 can achieve Figure 1 The various processes in the method embodiment achieve the same beneficial effects and are not described again here to avoid repetition.

[0164] The embodiment of the present invention also provides a server. Figure 5 The server may include a processor 901, a memory 902, and a program 9021 stored in the memory 902 and executable on the processor 901. When the program 9021 is executed by the processor 901, Figure 1 Any steps in the corresponding method embodiments and achieving the same beneficial effects will not be repeated here.

[0165] A person skilled in the art will understand that all or part of the steps of the above-mentioned embodiment method can be completed by hardware related to program instructions, and the program can be stored in a readable medium. The embodiment of the present invention also provides a readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned Figure 1-Figure 2 Any steps in the corresponding method embodiments can achieve the same technical effects and will not be described again here to avoid repetition.

[0166] The storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0167] The above is a preferred implementation of the embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An information push method, characterized in that: The method comprises: Get the selected push information; Performing event extraction on the to-be-selected push information to obtain a key event and a first target sentence related to the key event in the to-be-selected push information; Filter the key events based on the user's historical behavior data to obtain target events; generating target information according to a first target sentence corresponding to the target event; Pushing the target information to the user; After acquiring the to-be-selected push information and before generating the target information according to the first target sentence corresponding to the target event, the method further includes: Obtaining the target importance of each sentence in the selected push information; Determining a second target sentence based on the target importance of each sentence in the selected push information, where the second target sentence is the top N target importance sentences sorted from high to low in order of importance, where N is a positive integer; Generating target information according to the first target sentence corresponding to the target event includes: generating the target information according to the second target sentence and the first target sentence corresponding to the target event; The selected push information is a news article.

2. The method according to claim 1, characterized in that The obtaining of the target importance of each sentence in the selected push information includes: Calculating the first importance of each sentence in the selected push information based on the semantic similarity between the sentences; determining the second importance of each sentence according to the similarity between each sentence and the title of the selected push information; determining a third importance of each sentence based on a similarity between each sentence and a keyword of the user, wherein the keyword of the user is a keyword of a browsing content preference determined based on the historical behavior data; A target importance of each sentence is determined based on the first importance, the second importance, and the third importance of each sentence.

3. The method according to claim 1, characterized in that The generating the target information according to the second target sentence and the first target sentence corresponding to the target event includes: generating an initial text based on the second target sentence and the first target sentence corresponding to the target event; Calculating a target similarity between a first sentence and a second sentence in the initial text, where the first sentence and the second sentence are any two sentences in the initial text; If the target similarity is greater than a first preset threshold, removing the first sentence or the second sentence from the initial text to obtain a target text; The target information is obtained according to the target text.

4. The method according to claim 3, characterized in that If the target similarity is greater than a first preset threshold, removing the first sentence or the second sentence from the initial text to obtain a target text includes: If the target similarity is greater than a first preset threshold, removing the first sentence or the second sentence from the initial text to obtain an intermediate text; Calculating a probability of a third sentence in the intermediate text, where the third sentence is any sentence in the intermediate text; If the probability is less than a second preset threshold, error correction is performed on the third sentence in the intermediate text to obtain the target text.

5. The method according to claim 2, characterized in that The calculating the first importance of each sentence in the selected push information based on the semantic similarity between the sentences includes: Splitting the to-be-selected push information to obtain multiple words; Obtain word vectors for the multiple words; Obtaining a sentence vector of a third target sentence according to the word vectors of each word in the third target sentence, wherein the third target sentence is any one of the sentences in the to-be-selected push information; determining, based on the sentence vectors of any two third target sentences in the to-be-selected push information, the semantic similarity of the any two third target sentences; The first importance of each third target sentence is determined according to the semantic similarity between any two third target sentences in the to-be-selected push information.

6. The method according to claim 1, characterized in that After pushing the target information to the user, the method further includes: Recording the current behavior data of the user operating the target information; The historical behavior data is updated using the current behavior data.

7. An information push device, characterized in that: include: The acquisition module is used to obtain the push information to be selected; An event extraction module, configured to extract events from the selected push information to obtain a key event and a first target sentence related to the key event in the selected push information; A screening module is used to screen the key events based on the user's historical behavior data to obtain target events; A generating module, configured to generate target information according to a first target sentence corresponding to the target event; A push module, configured to push the target information to the user; The information pushing device further includes: A second acquisition module is used to obtain the target importance of each sentence in the selected push information; a determination module, configured to determine a second target sentence based on the target importance of each sentence in the candidate push information, wherein the second target sentence is the sentences corresponding to the top N target importances sorted from high to low according to the target importance of each sentence in the candidate push information, where N is a positive integer; The generating module is configured to generate the target information according to the second target sentence and the first target sentence corresponding to the target event; The selected push information is a news article.

8. A server comprising: A transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; wherein the processor is configured to read the program in the memory to implement the steps of the information push method as described in any one of claims 1 to 6.

9. A readable storage medium for storing a program, characterized in that: When the program is executed by a processor, the steps of the information push method according to any one of claims 1 to 6 are implemented.

Images