Method and device for mining user potential preferences

Abstract

The present invention provides a method and device for mining user potential preferences, the method comprising: preprocessing the acquired user text data to obtain a sentence set S and a word set W; generating multiple word association graphs based on high-frequency words in the word set W; determining low-frequency candidate words based on the multiple word association graphs and the sentence set S; adding the low-frequency candidate words to the multiple word association graphs; and determining potential preference events based on the co-occurrence rate between the low-frequency candidate words and each high-frequency word. The method and device for mining user potential preferences provided by the present invention mine the user's potential interests in the user's intention, find out excellent resources that are related to the user's intention but far away, mine classic and essential resources that are poorly related to the user's interest and far away, meet the user's potential interests, and analyze implicit associations, thereby creating diversity and novelty for the user, and further solving the long-tail problem in big data resources.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for mining potential preferences of users. Background Art

[0002] User potential preferences refer to the rational and biased choices made by users when considering goods and services. This is the comprehensive result of user cognition, psychological feelings and rational economic trade-offs. In daily life, people often use their preferences to assist in daily decision-making, or simply use preferences to express preferences and biased opinions without involving actual choices. However, in a business environment, it is crucial for companies to tap into users' potential preferences.

[0003] There are various methods for mining users' potential preferences, including traditional carpet search and advertising search methods, as well as key clue mining and decoupled representation learning that focus more on data analysis and technology applications.

[0004] However, existing methods for mining users' potential preferences have the problem of the long tail in big data resources. The long tail problem in big data resources mainly involves the imbalance of data distribution, that is, minority class samples have a large number of samples (head class), while other classes have only a small number of samples (tail class). This causes the model to be more biased towards head data and perform poorly on tail data during deep learning training, thus causing trouble for the learning process. Summary of the invention

[0005] The present invention provides a method and device for mining potential user preferences, which are used to solve the long tail problem in big data resources in the prior art methods for mining potential user preferences.

[0006] The present invention provides a method for mining user potential preferences, comprising:

[0007] Preprocess the acquired user text data to obtain a sentence set S and a word set W;

[0008] Based on the high-frequency words in the word set W, generating multiple word association graphs;

[0009] Determine low-frequency candidate words based on the multiple word association graphs and the sentence set S;

[0010] Adding the low-frequency candidate words to the plurality of word association graphs;

[0011] Based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word, the user's potential preference event is determined.

[0012] In some embodiments, the preprocessing of the acquired user text data includes:

[0013] Perform word segmentation and segmentation processing on the acquired user text data to obtain multiple words and multiple short sentences;

[0014] Using a stop word dictionary to perform stop word processing on the multiple words and multiple short sentences to eliminate meaningless words;

[0015] Use a synonym dictionary to merge multiple words and multiple short sentences after stop word processing.

[0016] In some embodiments, the generating of multiple word association graphs based on the high-frequency words in the word set W includes:

[0017] The words in the word set W whose frequency is greater than a preset threshold value θ are regarded as high-frequency words, and a high-frequency word set N is generated from the high-frequency words;

[0018] Calculate the co-occurrence rate of every two high-frequency words in the high-frequency word set N in the same sentence;

[0019] Multiple word association graphs are generated by taking word pairs whose co-occurrence rate is greater than a preset threshold μ as nodes and solid lines between the word pairs as edges.

[0020] In some embodiments, determining low-frequency candidate words based on the multiple word association graphs and the sentence set S includes:

[0021] Based on the multiple word association graphs and the sentence set S, determining an association preference value key(w) of each word w in the word set W, wherein the association preference value key(w) represents a conditional probability of the user using word w based on all intent group sets; wherein each word association graph corresponds to an intent group;

[0022] The words whose associated preference value key(w) is greater than a preset threshold δ are determined as low-frequency candidate words.

[0023] In some embodiments, determining the user's potential preference event based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word includes:

[0024] Calculate the co-occurrence rate between each low-frequency candidate word and each high-frequency word;

[0025] The low-frequency candidate words whose co-occurrence rate is greater than the preset threshold ε are regarded as low-frequency words, and form high-low frequency word pairs with the corresponding high-frequency words;

[0026] Based on the high- and low-frequency word pairs, potential preference events of the user are determined.

[0027] In some embodiments, determining the user's potential preference events based on the high- and low-frequency word pairs includes:

[0028] Taking the high-frequency and low-frequency word pairs as nodes and the dotted lines between the high-frequency and low-frequency word pairs as edges, updating the word association graph;

[0029] For each high-frequency word in the updated word association graph, count the number of dotted lines connected to each high-frequency word;

[0030] Determine the high-frequency words whose number of dotted lines connected to each high-frequency word is greater than a preset threshold σ as keywords of the user's main idea;

[0031] The high-frequency words at one end of the keyword are determined as the user's main intention event, and the low-frequency words at the other end of the keyword are determined as the user's potential preference event.

[0032] The present invention also provides a user potential preference mining device, comprising:

[0033] A preprocessing module is used to preprocess the acquired user text data to obtain a sentence set S and a word set W;

[0034] A word association graph construction module, used to generate multiple word association graphs based on high-frequency words in the word set W;

[0035] A low-frequency candidate word determination module, used to determine low-frequency candidate words based on the multiple word association graphs and the sentence set S;

[0036] a preference association determination module, configured to add the low-frequency candidate words to the plurality of word association graphs;

[0037] The potential preference mining module is used to determine the user's potential preference events based on the co-occurrence rate between the low-frequency candidate words and each high-frequency word.

[0038] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the above-mentioned methods for mining potential user preferences is implemented.

[0039] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for mining potential user preferences as described in any one of the above is implemented.

[0040] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for mining potential user preferences as described above is implemented.

[0041] The method and device for mining user potential preferences provided by the present invention mine the user's potential interests in the user's intention, find out the excellent resources that are related to the user's intention but far away, mine the classic and essential resources that are poorly related to the user's interest and far away, meet the user's potential interests, and analyze implicit associations, thereby creating diversity and novelty for the user, and further solving the long-tail problem in big data resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 It is a flow chart of the method for mining user potential preferences provided by the present invention.

[0044] Figure 2 It is a structural schematic diagram of a device for mining user potential preferences provided by the present invention.

[0045] Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Figure 1 is a flow chart of the method for mining user potential preferences provided by the present invention, such as Figure 1 As shown, the method includes the following:

[0048] Step 101: pre-process the acquired user text data to obtain a sentence set S and a word set W.

[0049] First of all, we need to clarify what kind of events are the user's potential preferences. In the text data set, mining the user's potential preferences is to find those rare and unpopular events that have important value and significance.

[0050] Therefore, user potential preferences must first be rare (low frequency) in the text dataset and closely related to the main content posted by the user (keywords). User preferences can be keywords, but not all keywords are potential preferences. Potential preferences have the characteristics of low frequency and rarity, so they should be data that is scarce in the dataset but related to user intentions.

[0051] For example, when using users’ Weibo posts as text datasets for potential preference mining, the definition of potential preferences must meet two conditions: first, the event should appear less frequently in the dataset; second, it should be closely related to the main points of the user’s posts.

[0052] Specifically, the preprocessing of the acquired user text data includes:

[0053] Perform word segmentation and segmentation processing on the acquired user text data to obtain multiple words and multiple short sentences;

[0054] Using a stop word dictionary to perform stop word processing on the multiple words and multiple short sentences to eliminate meaningless words;

[0055] Use a synonym dictionary to merge multiple words and multiple short sentences after stop word processing.

[0056] For example, the text data of Weibo is an unstructured data type and needs to be processed by natural language processing technology for word segmentation and segmentation. After segmentation, each Weibo text data will contain many meaningless words such as conjunctions and function words. This article uses the "stop word dictionary" to remove those meaningless words.

[0057] At the same time, this algorithm also uses the "synonym dictionary" to merge synonyms. For English words, rooting is done, and multiple tenses, singulars and plurals of the words are rooted into the general singular form; for Chinese words, semantic merging is done, and different words are merged if they represent the same meaning.

[0058] In addition to extracting words, short sentences are also required. The text preprocessing stage finally outputs two sets: sentence set S and word set W. At the same time, the frequency of each word in each document is recorded as a basic indicator for the next step of constructing intent groups.

[0059] Step 102: Generate multiple word association graphs based on the high-frequency words in the word set W.

[0060] Specifically, the generating of a plurality of word association graphs based on the high-frequency words in the word set W includes:

[0061] The words in the word set W whose frequency is greater than a preset threshold value θ are regarded as high-frequency words, and a high-frequency word set N is generated from the high-frequency words;

[0062] Calculate the co-occurrence rate of every two high-frequency words in the high-frequency word set N in the same sentence;

[0063] Multiple word association graphs are generated by taking word pairs whose co-occurrence rate is greater than a preset threshold μ as nodes and solid lines between the word pairs as edges.

[0064] For example, after text preprocessing, the words with a frequency greater than the threshold θ are selected as high-frequency words in the word set W, represented by black nodes in the figure, and the set is N. Then, the co-occurrence rate of each two high-frequency words in the set N in the same sentence is calculated, and the word pairs with a co-occurrence rate greater than the threshold µ are connected with solid lines in the figure. The expression for defining the co-occurrence rate R is as follows:

[0065]

[0066] Among them, n i and n j are all elements in the high-frequency word set N, q(n i ) represents n i The number of times it appears in sentence s, q(n i ∧n j ) represents n i and n j The number of times they appear in the same sentence.

[0067] The result of constructing the word association graph is to generate multiple intent groups. There are many keywords in the intent group, which represent the author's main point of view and are composed of high-frequency words.

[0068] Step 103: Determine low-frequency candidate words based on the multiple word association graphs and the sentence set S.

[0069] Specifically, determining low-frequency candidate words based on the multiple word association graphs and the sentence set S includes:

[0070] Based on the multiple word association graphs and the sentence set S, determining an association preference value key(w) of each word w in the word set W, wherein the association preference value key(w) represents a conditional probability of the user using word w based on all intent group sets; wherein each word association graph corresponds to an intent group;

[0071] The words whose associated preference value key(w) is greater than a preset threshold δ are determined as low-frequency candidate words.

[0072] For example, all the words appearing in the word association graph are from the high-frequency word set N. The high-frequency keywords represent the user's main preference. The embodiment of the present application further seeks to find those potential preference words that have low frequency of occurrence but are of great significance. Such words are words that can bind together the user's main ideas, that is, to mine those low-frequency words that have a high probability of co-occurring with the intent group representing the user's point of view.

[0073] For each word w in the set W, the associated preference value key(w) is calculated. This value is a real number between 0 and 1, which measures the probability that word w will appear if the user considers all the word clusters in the graph. The expression of the associated preference value key(w) is as follows:

[0074]

[0075] Among them, w represents a word in the word set W; G represents an intent group; g represents a group in the intent group G; s represents each sentence in the sentence set S; and q represents the frequency.

[0076] The Based() function indicates how many times the word w appears in the document based on the basic idea represented by a certain intent group g. More specifically, it indicates the co-occurrence between the word w and the phrase g, indicating the number of times the word w and g co-occur in the sentence s, that is, the number of times w appears in the sentence as a word in g;

[0077] The neighbors() function indicates the number of times a word in any sentence in the sentence set S includes a word in g.

[0078] key(w) represents the conditional probability of using word w given that the user is based on all intent clusters. Here, based divided by neighbors represents the occurrence rate of w in the neighborhood of the word in g, which means that if the user is interested in the main idea represented by a certain intent group, the conditional probability g of using word w is.

[0079] The words whose key(w) value is greater than the threshold δ are called low-frequency candidate words, and these words form the set H.

[0080] Step 104: Add the low-frequency candidate words to the multiple word association graphs.

[0081] Specifically, the words in set H are added to the intent group and represented by red nodes. If the word already appears in the graph, no operation is performed (i.e., high-frequency words are excluded).

[0082] Step 105: Determine the user's potential preference events based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word.

[0083] Specifically, determining the user's potential preference event based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word includes:

[0084] Calculate the co-occurrence rate between each low-frequency candidate word and each high-frequency word;

[0085] The low-frequency candidate words whose co-occurrence rate is greater than the preset threshold ε are regarded as low-frequency words, and form a high-low frequency word pair with the high-frequency words used in calculating the co-occurrence rate;

[0086] Based on the high- and low-frequency word pairs, potential preference events of the user are determined.

[0087] Specifically, determining the user's potential preference events based on the high- and low-frequency word pairs includes:

[0088] Taking the high-frequency and low-frequency word pairs as nodes and the dotted lines between the high-frequency and low-frequency word pairs as edges, updating the word association graph;

[0089] For each high-frequency word in the updated word association graph, count the number of dotted lines connected to each high-frequency word;

[0090] Determine the high-frequency words whose number of dotted lines connected to each high-frequency word is greater than a preset threshold σ as keywords of the user's main idea;

[0091] The high-frequency words at one end of the keyword are determined as the user's main intention event, and the low-frequency words at the other end of the keyword are determined as the user's potential preference event.

[0092] For example, for a low-frequency candidate word added to the graph, the co-occurrence rate between the word and the high-frequency word is calculated, and a threshold ε is given. When the co-occurrence rate is greater than the threshold ε, a dotted line is used to connect the low-frequency candidate word with the high-frequency word in the graph to form a preference association.

[0093] For the nodes connected with dotted lines in the figure, they are arranged in descending order according to the number of dotted lines connected to them. Then, according to a given threshold σ, all nodes connected with dotted lines above the threshold σ are selected as keywords of the user's main purpose, among which the words represented by the black nodes are the user's main intention events, and the words represented by the red nodes connected with the dotted lines are potential preference events.

[0094] The user potential preference mining method provided by the present invention mines the user's potential interests in the user's intention, finds out the excellent resources that are related to the user's intention but are far away, mines the classic essence resources that are poorly related to the user's interest and are far away, satisfies the user's potential interests, and analyzes implicit associations, thereby creating diversity and novelty for the user, and further solving the long-tail problem in big data resources.

[0095] The user potential preference mining device provided by the present invention is described below. The user potential preference mining device described below and the user potential preference mining method described above can be referenced to each other.

[0096] Figure 2 is a schematic diagram of the structure of the user potential preference mining device provided by the present invention, such as Figure 2 As shown, the present invention provides a user potential preference mining device, comprising:

[0097] The preprocessing module 201 is used to preprocess the acquired user text data to obtain a sentence set S and a word set W; the word association graph construction module 202 is used to generate multiple word association graphs based on the high-frequency words in the word set W; the low-frequency candidate word determination module 203 is used to determine the low-frequency candidate words based on the multiple word association graphs and the sentence set S; the preference association determination module 204 is used to add the low-frequency candidate words to the multiple word association graphs; the potential preference mining module 205 is used to determine the user's potential preference events based on the co-occurrence rate between the low-frequency candidate words and each high-frequency word.

[0098] In some embodiments, the preprocessing of the acquired user text data includes:

[0099] Perform word segmentation and segmentation processing on the acquired user text data to obtain multiple words and multiple short sentences;

[0100] Using a stop word dictionary to perform stop word processing on the multiple words and multiple short sentences to eliminate meaningless words;

[0101] Use a synonym dictionary to merge multiple words and multiple short sentences after stop word processing.

[0102] In some embodiments, the generating of multiple word association graphs based on the high-frequency words in the word set W includes:

[0103] The words in the word set W whose frequency is greater than a preset threshold value θ are regarded as high-frequency words, and a high-frequency word set N is generated from the high-frequency words;

[0104] Calculate the co-occurrence rate of every two high-frequency words in the high-frequency word set N in the same sentence;

[0105] Multiple word association graphs are generated by taking word pairs whose co-occurrence rate is greater than a preset threshold μ as nodes and solid lines between the word pairs as edges.

[0106] In some embodiments, determining low-frequency candidate words based on the multiple word association graphs and the sentence set S includes:

[0107] Based on the multiple word association graphs and the sentence set S, determining an association preference value key(w) of each word w in the word set W, wherein the association preference value key(w) represents a conditional probability of the user using word w based on all intent group sets; wherein each word association graph corresponds to an intent group;

[0108] The words whose associated preference value key(w) is greater than a preset threshold δ are determined as low-frequency candidate words.

[0109] In some embodiments, determining the user's potential preference event based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word includes:

[0110] Calculate the co-occurrence rate between each low-frequency candidate word and each high-frequency word;

[0111] The low-frequency candidate words whose co-occurrence rate is greater than the preset threshold ε are regarded as low-frequency words, and form high-low frequency word pairs with the corresponding high-frequency words;

[0112] Based on the high- and low-frequency word pairs, potential preference events of the user are determined.

[0113] In some embodiments, determining the user's potential preference events based on the high- and low-frequency word pairs includes:

[0114] Taking the high-frequency and low-frequency word pairs as nodes and the dotted lines between the high-frequency and low-frequency word pairs as edges, updating the word association graph;

[0115] For each high-frequency word in the updated word association graph, count the number of dotted lines connected to each high-frequency word;

[0116] Determine the high-frequency words whose number of dotted lines connected to each high-frequency word is greater than a preset threshold σ as keywords of the user's main idea;

[0117] The high-frequency words at one end of the keyword are determined as the user's main intention event, and the low-frequency words at the other end of the keyword are determined as the user's potential preference event.

[0118] Specifically, the above-mentioned user potential preference mining device provided in the embodiment of the present application can implement all the method steps implemented in the above-mentioned user potential preference mining method embodiment, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.

[0119] Figure 3 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 3As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the user potential preference mining method, which includes:

[0120] Preprocess the acquired user text data to obtain a sentence set S and a word set W;

[0121] Based on the high-frequency words in the word set W, generating multiple word association graphs;

[0122] Determine low-frequency candidate words based on the multiple word association graphs and the sentence set S;

[0123] Adding the low-frequency candidate words to the plurality of word association graphs;

[0124] Based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word, the user's potential preference event is determined.

[0125] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0126] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the user potential preference mining method provided by the above methods, the method includes:

[0127] Preprocess the acquired user text data to obtain a sentence set S and a word set W;

[0128] Based on the high-frequency words in the word set W, generating multiple word association graphs;

[0129] Determine low-frequency candidate words based on the multiple word association graphs and the sentence set S;

[0130] Adding the low-frequency candidate words to the plurality of word association graphs;

[0131] Based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word, the user's potential preference event is determined.

[0132] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the user potential preference mining method provided by the above methods, the method comprising:

[0133] Preprocess the acquired user text data to obtain a sentence set S and a word set W;

[0134] Based on the high-frequency words in the word set W, generating multiple word association graphs;

[0135] Determine low-frequency candidate words based on the multiple word association graphs and the sentence set S;

[0136] Adding the low-frequency candidate words to the plurality of word association graphs;

[0137] Based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word, the user's potential preference event is determined.

[0138] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0139] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for mining user potential preferences, characterized in that: include: Preprocess the acquired user text data to obtain a sentence set S and a word set W; Based on the high-frequency words in the word set W, generating multiple word association graphs; Determine low-frequency candidate words based on the multiple word association graphs and the sentence set S; Adding the low-frequency candidate words to the plurality of word association graphs; Determine the user's potential preference event based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word; The determining of the user's potential preference event based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word includes: Calculate the co-occurrence rate between each low-frequency candidate word and each high-frequency word; The low-frequency candidate words whose co-occurrence rate is greater than the preset threshold ε are regarded as low-frequency words, and form high-low frequency word pairs with the corresponding high-frequency words; Based on the high- and low-frequency word pairs, determining the user's potential preference events; The determining of the user's potential preference events based on the high- and low-frequency word pairs includes: Taking the high-frequency and low-frequency word pairs as nodes and the dotted lines between the high-frequency and low-frequency word pairs as edges, updating the word association graph; For each high-frequency word in the updated word association graph, count the number of dotted lines connected to each high-frequency word; Determine the high-frequency words whose number of dotted lines connected to each high-frequency word is greater than a preset threshold σ as keywords of the user's main idea; The high-frequency words at one end of the keyword are determined as the user's main intention event, and the low-frequency words at the other end of the keyword are determined as the user's potential preference event.

2. The method for mining user potential preferences according to claim 1, characterized in that: The preprocessing of the acquired user text data includes: Perform word segmentation and segmentation processing on the acquired user text data to obtain multiple words and multiple short sentences; Using a stop word dictionary to perform stop word processing on the multiple words and multiple short sentences to eliminate meaningless words; Use a synonym dictionary to merge multiple words and multiple short sentences after stop word processing.

3. The method for mining user potential preferences according to claim 1, characterized in that: The generating of a plurality of word association graphs based on the high frequency words in the word set W includes: The words in the word set W whose frequency is greater than a preset threshold value θ are regarded as high-frequency words, and a high-frequency word set N is generated from the high-frequency words; Calculate the co-occurrence rate of every two high-frequency words in the high-frequency word set N in the same sentence; Multiple word association graphs are generated by taking word pairs whose co-occurrence rate is greater than a preset threshold μ as nodes and solid lines between the word pairs as edges.

4. The method for mining user potential preferences according to claim 1, characterized in that: The determining of low-frequency candidate words based on the multiple word association graphs and the sentence set S includes: Based on the multiple word association graphs and the sentence set S, determining an association preference value key(w) of each word w in the word set W, wherein the association preference value key(w) represents a conditional probability of the user using word w based on all intent group sets; wherein each word association graph corresponds to an intent group; The words whose associated preference value key(w) is greater than a preset threshold δ are determined as low-frequency candidate words.

5. A device for mining user potential preferences, characterized in that: include: A preprocessing module is used to preprocess the acquired user text data to obtain a sentence set S and a word set W; A word association graph construction module, used to generate multiple word association graphs based on high-frequency words in the word set W; A low-frequency candidate word determination module, used to determine low-frequency candidate words based on the multiple word association graphs and the sentence set S; a preference association determination module, configured to add the low-frequency candidate words to the plurality of word association graphs; A potential preference mining module, for determining a user's potential preference events based on the co-occurrence rate between the low-frequency candidate words and each high-frequency word; The determining of the user's potential preference event based on the co-occurrence rate between the low-frequency candidate word and each high-frequency word includes: Calculate the co-occurrence rate between each low-frequency candidate word and each high-frequency word; The low-frequency candidate words whose co-occurrence rate is greater than the preset threshold ε are regarded as low-frequency words, and form high-low frequency word pairs with the corresponding high-frequency words; Based on the high- and low-frequency word pairs, determining the user's potential preference events; The determining of the user's potential preference events based on the high- and low-frequency word pairs includes: Taking the high-frequency and low-frequency word pairs as nodes and the dotted lines between the high-frequency and low-frequency word pairs as edges, updating the word association graph; For each high-frequency word in the updated word association graph, count the number of dotted lines connected to each high-frequency word; Determine the high-frequency words whose number of dotted lines connected to each high-frequency word is greater than a preset threshold σ as keywords of the user's main idea; The high-frequency words at one end of the keyword are determined as the user's main intention event, and the low-frequency words at the other end of the keyword are determined as the user's potential preference event.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for mining potential user preferences as described in any one of claims 1 to 4 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for mining user potential preferences as described in any one of claims 1 to 4 is implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for mining user potential preferences as described in any one of claims 1 to 4 is implemented.

Images