Method and device for recommending bidding words based on theme recognition, and storage medium

Abstract

The application provides a bidding word recommendation method based on theme recognition, which comprises the following steps: S1, obtaining a trained theme recognition model; S2, inputting bidding words and search words of a theme type to be recognized into the trained theme recognition model to recognize the theme type of the bidding words and the search words to be recognized; S3, determining, according to the opening rate of the bidding words and the search words of the theme type to be recognized and the recognized theme type, the bidding word opening rate of each theme in the recognized theme type for the bidding words, and determining the search word opening rate of each theme in the recognized theme type for the search words; and S4, for a bidding word theme with a theme opening rate of 0, obtaining search words with openings in search words corresponding to the theme and the bidding word theme, and recommending the search words as bidding word extension words. The above technical scheme can improve the pertinence of company bidding extension words and improve the use efficiency of bidding words.

Description

Technical Field

[0001] This invention relates to the field of online advertising, and in particular to a method, apparatus, and storage medium for recommending bidding keywords based on topic recognition. Background Technology

[0002] Pay-per-click (PPC) advertising is a method of placing ads on internet search engines and content websites. The basic principle is that advertisers bid on keywords, and when users use relevant keywords in search engines, these ads appear in the relevant search results. When users click on these ads, the advertiser pays a fee. Therefore, the precision of the keywords bid in PPC helps achieve a higher return on investment (ROI).

[0003] However, the industry currently relies mainly on keyword tools to continuously add more keywords based on internet search engine popularity. This method is not very targeted and cannot be optimized based on historical performance feedback. This leads to an ever-increasing number of bidding keywords, resulting in higher management costs, lower keyword effectiveness, and thus significant waste in bidding. Summary of the Invention

[0004] The embodiments of the present invention provide a method, apparatus and storage medium for recommending bidding keywords based on topic recognition, in order to solve the problems of weak targeting and inefficient use of bidding keywords in current companies.

[0005] On the one hand, a bidding keyword recommendation method based on topic recognition is provided, including the following steps:

[0006] S1, obtaining the trained topic recognition model, including:

[0007] S11, extract keywords from multiple bidding terms and search terms used to determine the topic category, and determine the topic category of the bidding terms and search terms;

[0008] S12, Based on the determined topic classification, label the multiple bidding terms and search terms used to train the topic recognition model with topic categories;

[0009] S13, A topic recognition model is constructed based on an LSTM neural network, and the topic recognition model is trained using bidding words and search words that have been labeled with topic categories;

[0010] S2, Input the bidding terms and search terms of the topic type to be identified into the trained topic identification model to identify the topic type of the bidding terms and search terms;

[0011] S3, based on the opening rates of the bidding terms and search terms of the topic type to be identified and the identified topic type, determine the opening rate of the bidding term for each topic in the identified topic type for the bidding term, and determine the opening rate of the search term for each topic in the identified topic type for the search term;

[0012] S4. For bidding keyword topics with an opening rate of 0, obtain the search terms with openings in the search terms corresponding to the bidding keyword topic, and recommend the search terms as bidding keyword expansion terms.

[0013] Preferably, in the method, step S11 includes:

[0014] The jieba word segmentation tool was used to segment multiple bidding keywords and search terms used to determine the topic category, and each word after segmentation was obtained.

[0015] Calculate the TF-IDF of each word after word segmentation to obtain the weight value of each word;

[0016] Ranked according to their TF-IDF values, with larger values ​​ranked higher, a predetermined number of words are selected from the top-ranked words to identify candidate topic words for the topic classification.

[0017] Preferably, in the method, step S13 includes:

[0018] The bidding keywords and search terms that have been categorized by topic are segmented into words;

[0019] The words obtained after word segmentation are encoded using word2vec;

[0020] The encoded data is then input into an LSTM neural network for training.

[0021] After training, the category with the highest probability is selected as the topic category using softmax.

[0022] Preferably, in the method, step S2 includes:

[0023] S21, each bid or search term in the bid or search terms of the topic type to be identified is vectorized using word2vec, and then input into the trained topic recognition model;

[0024] S22, The LSTM neural network in the trained topic recognition model performs semantic extraction on the vectors corresponding to bidding words or search words to obtain the features of bidding words;

[0025] S23, through a tight network, output the features of bidding terms or search terms into an output vector with a length equal to the number of topic categories;

[0026] S24, calculate the probability of the corresponding bidding keyword or search term on each topic using softmax;

[0027] S25, select the topic with the highest probability as the topic of the corresponding bidding keyword or search term.

[0028] Preferably, in the method, the aperture ratio in step S3 is calculated using the following formula:

[0029] Opening ratio = Number of openings / Number of traffic connections

[0030] The number of conversations refers to the number of actual visitors who have a dialogue with customer service and who have said at least one sentence.

[0031] Preferably, in the method, in step S3,

[0032] Open rate of search term topic = Number of open links for the search term topic / Number of traffic accesses corresponding to the search term topic

[0033] The open rate of a bidding keyword topic = the number of open keywords for that bidding keyword topic / the number of traffic accesses corresponding to that bidding keyword topic.

[0034] Preferably, in the method, in step S4, the search term corresponding to the bidding keyword topic is the same as the bidding keyword topic.

[0035] On the other hand, a topic recognition-based bidding keyword recommendation device is provided, including a memory and a processor, wherein the memory stores at least one program, which is executed by the processor to implement any of the methods described above.

[0036] In another aspect, a computer-readable storage medium is provided, wherein at least one program is stored therein, the at least one program being executed by a processor to implement any of the methods described above.

[0037] The above technical solution has the following technical effects:

[0038] The technical solution proposed in this invention provides a topic-based keyword recommendation method for bidding keywords, which enables intelligent keyword expansion. It trains a domain topic recognition model based on bidding keywords and search terms, automatically categorizes the topics of enterprise bidding keywords and user search terms, and uses topic and user open-ended search terms analysis to recommend open-ended search terms for bidding keyword topics with zero open-ended search terms as bidding keyword expansion. This solves the problem of weak targeting and inefficiency in companies' blind bidding keyword expansion, and improves the targeting and effectiveness of bidding keywords with zero open-ended search terms. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating a topic-based keyword recommendation method according to an embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram illustrating the principle of a topic-based keyword recommendation method according to another embodiment of the present invention.

[0041] Figure 3 This is a schematic diagram of a bidding keyword recommendation device based on topic recognition according to an embodiment of the present invention. Detailed Implementation

[0042] To further illustrate the various embodiments, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, primarily used to illustrate the embodiments and to explain the operating principles of the embodiments in conjunction with the relevant descriptions in the specification. With reference to these drawings, those skilled in the art should be able to understand other possible implementations and the advantages of the present invention. Components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0043] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments.

[0044] Example 1:

[0045] Figure 1 This is a flowchart illustrating a topic-based keyword recommendation method according to an embodiment of the present invention. Figure 1 The method in this embodiment includes the following steps:

[0046] S1, obtaining a trained topic recognition model, including: S11, extracting keywords from multiple bidding terms and search terms used to determine topic classification, and determining the topic classification of the bidding terms and search terms; S12, labeling multiple bidding terms and search terms used to train the topic recognition model with topic categories according to the determined topic classification; S13, constructing a topic recognition model based on an LSTM neural network, and training the topic recognition model using bidding terms and search terms with topic category labels;

[0047] S2, Input the bidding terms and search terms of the topic type to be identified into the trained topic identification model to identify the topic type of the bidding terms and search terms;

[0048] S3, based on the open rates of the bidding terms and search terms of the topic type to be identified and the identified topic type, determine the open rate of the bidding terms for each topic in the identified topic type for the bidding terms, and the open rate of the search terms for each topic in the identified topic type for the search terms;

[0049] S4. For bidding keyword topics with an open rate of 0, obtain search terms with open rates that are part of the same topic as the bidding keyword topic, and recommend these search terms as bidding keyword expansion terms.

[0050] Example 2:

[0051] Figure 2 This is a schematic diagram illustrating the principle of a topic-based keyword recommendation method according to another embodiment of the present invention. Figure 2 The overall process of implementing the method in this embodiment of the invention can be seen. The prerequisite for implementing the technical solution of this embodiment is the availability of bidding keywords, search terms, and the corresponding user responses to these bidding and search terms. Generally, bidding keywords are the keywords or phrases that a company bids for on the platform; search terms are the words or phrases entered by users as visitors in the search box of the platform or search engine; "speaking" refers to a user clicking on a corresponding advertisement after searching, entering online customer service, generating a dialogue, and engaging in communication. Usually, speaking at least one sentence is considered as engaging in communication. Table 1 provides examples of the records generated by visitor visits.

[0052]

[0053] Table 1

[0054] The steps of this embodiment of the present invention will be described in detail below.

[0055] Step one: Extract keywords from bidding terms and search terms to determine the theme category, which is to determine the project category.

[0056] The bidding keywords (company bidding keywords) and search terms (user search terms) are segmented using the jieba word segmentation tool. The TF-IDF of each segmented keyword is calculated to obtain its weight value. Keywords are ranked according to their TF-IDF values, with higher values ​​ranking higher. A predetermined number of keywords with the highest rankings are selected as candidate keywords for topic classification. For example, the top 100 keywords with the highest TF-IDF values ​​are selected as candidate keywords. These candidate keywords might include: myopia, femtosecond LASIK, laser surgery, improving vision, blurred vision, and ophthalmology hospitals.

[0057] Step two: Based on the topic classification, label the bidding keywords and search terms with topic categories, i.e., project categories. Table 2 shows an example of category labeling.

[0058]

[0059]

[0060] Table 2

[0061] Step 3: Construct a topic recognition model based on an LSTM neural network, and train the model using the bidding terms and search terms used for topic category labeling. Training the topic recognition model mainly includes:

[0062] 1. Segment the text into words. Here, "text" refers to the bidding terms and search terms used for training, and the result is the individual words after segmentation.

[0063] 2. Encode each word obtained from the above word segmentation using word2vec;

[0064] 3. Input the obtained encoding into the LSTM neural network for training to obtain the trained vector; specifically, during the training process, semantic features are extracted through the LSTM neural network, and the extracted features are output into the output vector through a dense network.

[0065] 4. The obtained trained vectors are passed through a softmax layer to select the category with the highest probability as the topic category of the corresponding bidding term or search term.

[0066] Step 4: Using the trained topic recognition model, identify the topic types of the bidding terms and search terms to be identified or to be identified;

[0067] This step builds upon the topic recognition model trained in step three above, predicting the topic for each bidding term and search term to be identified. For example, but not as a limitation, for the bidding term "how to reduce eye swelling," the topic recognition result is "eye swelling." Further, the openness of the corresponding bidding term and search term is obtained, i.e., whether it is open. The results for the bidding term "how to reduce eye swelling" are shown in Table 3.

[0068] How to reduce eye puffiness Swollen eyes no

[0069] Table 3

[0070] For example, this step specifically includes:

[0071] Based on the topic recognition model trained in step three, for example, the company's purchased bidding keywords for this month are vectorized sentence by sentence using word2vec. The obtained vectors are then input into the trained topic recognition model. This model extracts semantic features from the bidding keywords using an LSTM neural network. Finally, the extracted features are processed through a dense network, and the features of the bidding keywords are output into an output vector `final_output` of length equal to the number of topic categories. The output vector is then processed using a softmax layer, i.e., `softmax(final_output)`, which calculates the probability `probs` of the current bidding keyword on each topic. The topic with the highest probability `max(probs)` is selected as the topic of that bidding keyword. Typically, each topic is identified by a pre-assigned ID, so the ID corresponding to the topic with the highest probability `max(probs)` is ultimately selected.

[0072] By applying the above steps to each bidding keyword, the topic categories of all bidding keywords can be identified. Similarly, using the same model, the topics of all search terms can be obtained through the same method.

[0073] Based on user access records, the opening rates of bidding keywords and search terms for different topics can be obtained, thus providing the opening rate for each topic in bidding keywords and search terms. Table 4 shows the opening rates of bidding keywords and search terms for some topics.

[0074] Shortsighted 9.56％ 10.61％ Strabismus 8.25％ 9.14％ Laser surgery 7.29％ 7.12％ Lacrimal duct diseases 0 6.5％ Swollen eyes 0 5.3％

[0075] Table 4

[0076] Step 5: Based on the open rates of the bidding keywords and search terms for the topic types to be identified, and the identified topic types, determine the open rate of the bidding keywords for each topic in the identified topic types, and determine the open rate of the search terms for each topic in the identified topic types.

[0077] The following example illustrates how to calculate the open rate for each topic in bidding keywords and search terms. Specifically, a visitor's visit generates a record, as shown in Table 5 below.

[0078]

[0079]

[0080] Table 5

[0081] By calculating the corresponding topics for the "company bidding keywords" and "user search terms" in Table 5 using the steps described above, we can obtain Table 6:

[0082]

[0083] Table 6

[0084] For Table 6, by statistically analyzing the data by search topic and bidding topic, we can obtain Table 7, which shows the opening status of each search topic, and Table 8, which shows the opening status of each bidding topic.

[0085] Wherein: the number of conversations refers to the number of actual visitors who have a dialogue with customer service and whose visitors have said at least one sentence; the conversation rate = number of conversations / number of traffic accesses; here, the number of traffic accesses is the number of clicks on the accessed webpage;

[0086] Open rate of search term topic = Number of open links for that search term topic / Number of traffic accesses corresponding to that search term topic

[0087]

[0088]

[0089] Table 7

[0090] The open rate of a bidding keyword topic = the number of open entries for that bidding keyword topic / the number of traffic accesses corresponding to that bidding keyword topic.

[0091] Swollen eyes 0 0 Shortsighted 86 8.5％ Laser surgery 10 7.3％ ...... ...... ...... ...... ...... ......

[0092] Table 8

[0093] Step 6: Based on the bidding keyword topics or items with a 0-opening rate (i.e., an opening rate of 0), expand the bidding keyword recommendations by including search terms with corresponding topics, such as the same topic, that have openings.

[0094] Specifically, a list of topics with an open rate of 0 can be output, as shown in Table 9. From the search term topics, search terms with corresponding topics are found, and those with open rates are recommended as keywords for expansion. As shown in Table 10, the search terms "What eye drops should I use for red and swollen eyes?" and "How can I reduce the swelling in my red and swollen eyes?" with open rates can be recommended as keywords for expansion.

[0095]

[0096]

[0097] Table 9

[0098] Swollen eyes yes What eye drops should I use for red and swollen eyes? Swollen eyes yes What can I do to reduce the swelling in my eyes?

[0099] Table 10

[0100] The method of this invention for recommending and expanding bidding keywords can automatically classify bidding keywords of companies or enterprises as well as user search terms. By analyzing the topic and user responses, it recommends corresponding search terms for bidding keyword topics with zero response rate as bidding keyword expansion, which solves the problem of weak targeting and inefficiency of companies blindly expanding bidding keywords, and improves the targeting and effectiveness of bidding keywords with zero response rate.

[0101] Example 3:

[0102] This invention also provides a bidding keyword recommendation device based on topic recognition, such as... Figure 3 As shown, the device includes a processor 301, a memory 302, a bus 303, and a computer program stored in the memory 302 and executable on the processor 301. The processor 301 includes one or more processing cores. The memory 302 is connected to the processor 301 via the bus 303. The memory 302 is used to store program instructions. When the processor executes the computer program, it implements the steps in the above-described method embodiment of Embodiment 1 of the present invention.

[0103] Furthermore, as an executable solution, the topic-based keyword recommendation device can be a computer unit, which can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer unit may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above-described computer unit structure is merely an example and does not constitute a limitation on the computer unit. It may include more or fewer components, or combine certain components, or use different components. For example, the computer unit may also include input / output devices, network access devices, buses, etc., and this embodiment of the invention does not limit this.

[0104] Furthermore, as an executable solution, the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the computer unit, connecting various parts of the entire computer unit via various interfaces and lines.

[0105] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the computer unit by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0106] Example 4:

[0107] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the methods described in the embodiments of the present invention.

[0108] If the modules / units integrated in the computer unit are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), and software distribution media, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0109] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A bidding keyword recommendation method based on topic recognition, characterized in that, Includes the following steps: S1, obtaining the trained topic recognition model, including: S11, Extract keywords from multiple bidding terms and search terms used to determine the topic category, and determine the topic category of the bidding terms and search terms; S12, Based on the determined topic classification, label the multiple bidding terms and search terms used to train the topic recognition model with topic categories; S13, A topic recognition model is constructed based on an LSTM neural network, and the topic recognition model is trained using bidding words and search words that have been labeled with topic categories; S2, Input the bidding terms and search terms of the topic type to be identified into the trained topic identification model to identify the topic type of the bidding terms and search terms; S3, based on the opening rates of the bidding terms and search terms of the topic type to be identified and the identified topic type, determine the opening rate of the bidding term for each topic in the identified topic type for the bidding term, and determine the opening rate of the search term for each topic in the identified topic type for the search term; The aperture ratio is calculated using the following formula: Opening ratio = Number of openings / Number of traffic connections; Among them, the number of conversations refers to the number of actual visitors who have had a dialogue with customer service and whose visitors have said at least one sentence. The opening rate of a search term topic = the number of openings in the search term topic / the number of traffic accesses corresponding to the search term topic; The open rate of a bidding keyword topic = the number of open keywords for that bidding keyword topic / the number of traffic accesses corresponding to that bidding keyword topic; S4. For bidding keyword topics with an opening rate of 0, obtain the search terms with openings in the search terms corresponding to the bidding keyword topic, and recommend the search terms as bidding keyword expansion terms.

2. The method according to claim 1, characterized in that, Step S11 includes: The jieba word segmentation tool was used to segment multiple bidding keywords and search terms used to determine the topic category, and each word after segmentation was obtained. Calculate the TF-IDF of each word after word segmentation to obtain the weight value of each word; Ranked according to their TF-IDF values, with larger values ​​ranked higher, a predetermined number of words are selected from the top-ranked words to identify candidate topic words for the topic classification.

3. The method according to claim 1, characterized in that, Step S13 includes: The bidding keywords and search terms that have been categorized by topic are segmented into words; The words obtained after word segmentation are encoded using word2vec; The encoded data is then input into an LSTM neural network for training. After training, the category with the highest probability is selected as the topic category using softmax.

4. The method according to claim 1, characterized in that, Step S2 includes: S21, each bid or search term in the bid or search terms of the topic type to be identified is vectorized using word2vec, and then input into the trained topic recognition model; S22, the LSTM neural network in the trained topic recognition model performs semantic extraction on the vectors corresponding to the bidding words or search words to obtain the features of the bidding words or search words; S23, through a tight network, output the features of bidding terms or search terms into an output vector with a length equal to the number of topic categories; S24, calculate the probability of the corresponding bidding keyword or search term on each topic through the softmax layer; S25, select the topic with the highest probability as the topic of the corresponding bidding keyword or search term.

5. The method according to claim 1, characterized in that, In step S4, the search term corresponding to the bidding keyword topic is the same as the bidding keyword topic.

6. A keyword recommendation device based on topic recognition, characterized in that, It includes a memory and a processor, the memory storing at least one program, the at least one program being executed by the processor to implement the method as claimed in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, The storage medium stores at least one program segment, which is executed by a processor to implement the method as described in any one of claims 1 to 5.

Images