Prediction program, information processing apparatus, and prediction method

Abstract

To appropriately detect keyword stuffing.SOLUTION: An input first character string is allocated to a block satisfying a predetermined condition, a probability that keyword stuffing is present in a second character string is predicted using a feature amount of each character of the second character string included in each block and a detector 112 for detecting keyword stuffing, and a center and a length of a keyword segment in the second character string are predicted when the probability is equal to or greater than a predetermined threshold value.SELECTED DRAWING: Figure 10

Description

[Technical Field]

[0001] The present invention relates to a prediction program, an information processing device, and a prediction method. [Background technology]

[0002] Information retrieval is the task of extracting sources containing the information necessary to answer a search query. Information retrieval has been gaining attention in recent years due to the popularity of Retrieval-Augmented Generation (RAG). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2018-77806 [Patent Document 2] Japanese Patent Application Publication No. 2020-46792 [Patent Document 3] U.S. Patent Publication No. 2007 / 0192309 [Patent Document 4] U.S. Publication No. 2023 / 0107493 Summary of the Invention [Problem to be solved by the invention]

[0004] Information retrieval can be compromised by keyword stuffing.

[0005] Keyword stuffing is an attack that uses a string of keywords to boost information search rankings. Because RAG calculates relevance using only the content of the text, keywords in the content may affect information search rankings.

[0006] In one aspect, the object is to appropriately detect keyword stuffing. [Means for solving the problem]

[0007] In one aspect, the prediction program causes a computer to perform the following process: assigning an input first character string to blocks that satisfy predetermined conditions; predicting the probability that keyword stuffing is present in the second character string using features of each character of a second character string included in each block and a detector for detecting keyword stuffing; and predicting the center and length of keyword segments in the second character string if the probability is equal to or greater than a predetermined threshold. [Effects of the Invention]

[0008] In one aspect, keyword stuffing can be appropriately detected. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 10 is a diagram illustrating keyword stuffing. [Figure 2] 10A and 10B are diagrams illustrating a detection process of keyword stuffing in a related example. [Figure 3] 3 is a diagram illustrating a problem in the detection process of keyword stuffing in the related example shown in FIG. 2. FIG. [Figure 4] 10A and 10B are diagrams illustrating an example of input / output description in the embodiment. [Figure 5] FIG. 1 is a diagram illustrating a backbone network structure according to an embodiment. [Figure 6] 10A and 10B are diagrams illustrating a global detection process of keyword stuffing in an embodiment. [Figure 7] FIG. 2 is a diagram illustrating details of a detector in the embodiment. [Figure 8] 10A and 10B are diagrams illustrating a correction process using local information for a global detection result in an embodiment. [Figure 9]10 is a flowchart illustrating a keyword stuffing prediction process according to an embodiment. [Figure 10] 10A and 10B are diagrams illustrating a keyword stuffing prediction process according to an embodiment. [Figure 11] FIG. 1 is a block diagram illustrating an example of a hardware configuration of an information processing device according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] [A] Related Examples FIG. 1 is a diagram illustrating keyword stuffing.

[0011] In FIG. 1, keyword stuffing detection technology is used in the sentence indicated by reference symbol A1 to detect an unnatural string of keywords as indicated by the dashed frame indicated by reference symbol A2.

[0012] FIG. 2 is a diagram illustrating the detection process of keyword stuffing in a related example.

[0013] In the example shown in FIG. 2, a DNN (Deep Neural Network) is used to predict whether each character is keyword stuffing or not using a naive detection method.

[0014] When a string v is input as shown in code B1, the kth character of the string v is input as shown in code B2. k The probability that is keyword stuffing is predicted to be (Equation 1).

number

[0015] The above (Equation 1) is the feature representation of the following (Equation 2) (in other words, the final layer of the feature extractor T).

number

[0016] If the following formula (3) is true, then the letter v k is considered keyword stuffing (or, in other words, a single character within keyword stuffing).

number

[0017] FIG. 3 is a diagram for explaining the problem of the keyword stuffing detection process in the related example shown in FIG.

[0018] As shown by symbols C1 and C2 in Fig. 3, the continuity of the output may not be guaranteed in the keyword stuffing detection process shown in Fig. 2. In symbol C1, an output 0 is sandwiched between successive outputs of 1, and in symbol C2, an output 1 is sandwiched between successive outputs of 0.

[0019] Keyword stuffing generally appears in a continuous range because it has the greatest impact on information retrieval rank. While post-processing can smooth the results, as shown in C3, it may not be possible to correct false positives or negatives with a large number of characters, as shown in C4.

[0020] [B] Embodiment Hereinafter, embodiments will be described with reference to the drawings. However, the embodiments shown below are merely examples, and are not intended to exclude various modifications or application of techniques not explicitly stated in the embodiments. In other words, the present embodiments can be implemented with various modifications within the scope of the spirit thereof. Furthermore, each figure does not intend to include only the components shown in the figure, but may include other components, etc.

[0021] [B-1] Program structure example FIG. 4 is a diagram for explaining an example of input / output description in the embodiment.

[0022] The overall structure F of the keyword stuffing prediction process in the embodiment is v L →y LIn the example shown in FIG. 4, L=40.

[0023] The following (Equation 4) represents a string consisting of L characters.

number

[0024] The following (Equation 5) is a sequence indicating whether or not there is keyword stuffing, and the keyword stuffing location is "1".

number

[0025] A keyword segment represents one unit of a keyword location, and FIG. 4 shows keyword segments #1 and #2.

[0026] The following (Equation 6) represents the se notation of the i-th segment (s(start): start point, e(end): end point).

number

[0027] The following (Equation 7) represents the cl notation (c (center): center, l (length): length) of the i-th segment.

number

[0028] Hereinafter, the notation of y will be interchangeable as shown in the following (Equation 8).

number

[0029] FIG. 5 is a diagram illustrating a backbone network structure according to an embodiment.

[0030] The backbone network T (in other words, the feature extractor) of the keyword stuffing prediction process in the embodiment is L →H L It is a DNN that extracts feature representations for each character from a string of characters, represented as follows:

[0031] The following (Equation 9) represents the feature expression for each character.

number

[0032] θ in Figure 5 T represents the weight of the backbone network.

[0033] FIG. 6 is a diagram illustrating a global detection process for keyword stuffing in the embodiment.

[0034] In FIG. 6, the output from the feature extractor 111 (T(v) in FIG. 6) indicated by the symbol D1 is input to the detector 112 (S(H1), S(H2), S(H3) in FIG. 6) indicated by the symbol D2.

[0035] The feature extractor 111 allocates the input first character string (in other words, character string v) to a block that satisfies a predetermined condition. The process of allocating to blocks may be performed such that at least a portion of the second character strings included in adjacent blocks overlap each other.

[0036] In the example shown in FIG. 6, the sequence is divided into blocks (blocks #1, #2, and #3 in FIG. 6), and the probability of keyword stuffing presence, center position, and length for each block are predicted.

[0037] The detector 112(S) performs detection processing using the following (Equation 10): K represents a block size. The block size K can be set to various values, and keyword stuffing of any length may be detected.

number

[0038] In the output of the detector 112 (S), p represents the probability that the "center" of the keyword stuffing exists in the block, and c and l represent the center (c) and length (l) of the predicted keyword segment. A block without a keyword center is p k ≒0.

[0039] In the output shown by reference symbol D3, the probability that a keyword stuffing center exists in the character string corresponding to block #1 is high, with center c1=13 and length l1=25.

[0040] That is, detector 112 predicts the probability that keyword stuffing is present in the second character string using the feature values ​​of each character in the second character string included in each block. If the probability is equal to or greater than a predetermined threshold, detector 112 predicts the center and length of the keyword segment in the second character string.

[0041] The probability prediction process may predict the probability that the true center of the keyword stuffing is present in the second string. As mentioned above, a keyword segment is a unit of keyword stuffing location. The detector 112 may be trained so that when a training sample consisting of a string containing keyword stuffing is given, it can predict the true positions (centers and lengths) of all keyword segments in the training sample using multiple blocks.

[0042] FIG. 7 is a diagram illustrating the details of the detector 112 in this embodiment.

[0043] The detector 112, which may be referred to as a segment module, is a DNN that predicts keyword stuffing on a block-by-block basis.

[0044] Detector 112(S) is H K →Perform the operation expressed as [0,1]×[0,1]×[0,1].

[0045] (K,W) represents (block size, sliding window size), and H k is the notation of the feature of the kth block, H1=(h1,…,h K ), H2=(h W ,…,h W+K ),... holds true.

[0046] Also, S(H k ;θ S )=(p,b,l) holds. P is the confidence that keyword stuffing exists, and b is the block start position a k (=((k-1)*W+1) / L) (i.e., c=∃a k +b), where l is the length.

[0047] In the training phase, the loss function L(θ T , θ S ) is minimized.

number

number

[0048] In the above (Equation 12), the first term represents the sum of the centroid error and the length error (which we want to make small), and the second term represents the probability that the keyword exists (the real number should be 1) and the probability that the keyword does not exist (the real number should be 0).

[0049] The following (Equation 13) represents the prediction result of block k.

number

[0050] τ(k) is the index set of the keyword segment that the kth block is responsible for. The centroid c of segment i i The block k to which belongs is responsible for detection. Strictly speaking, k satisfies the following (Equation 14).

number

[0051] In the example shown in FIG. 7, τ(2)={1}, τ(k≠2)={} (empty set).

[0052] The following (Equation 15) represents the weight for a block in which a keyword segment exists, and the following (Equation 16) represents the weight for a block in which a keyword segment does not exist (negative block).

number

number

[0053] FIG. 8 is a diagram illustrating a correction process using local information for a global detection result in the embodiment.

[0054] The corrector 113 (R(h) in FIG. 8) corrects the boundaries of the keyword segments in the tentative prediction, which is the global detection result shown in E1, using information about the vicinity of the boundaries to obtain the final prediction shown in E2. The correction of the boundaries of the keyword segments may be performed using only information about the vicinity of the boundaries.

[0055] When M is the neighborhood width (an odd number), the corrector 113 performs processing based on the following (Equation 17).

number

[0056] In the example shown in FIG. 8, the neighborhood width M=7, and the corrector 113 (R(h 10:16 )) and a corrector 113 (R(h 35:41 )) is shown. n represents the number of characters from the left end of the string. R(h 10:16 ) is adjusted by +2 (2 towards the end of the string).35:41 ) is adjusted by -1 (1 towards the start of the string).

[0057] The corrector 113 performs training based on the residual error of the segment boundary in the training data. When the start point and the end point are at both ends of the keyword segment (the nth character), the residual error Error is expressed as Error = (true start point - corrected start point). 2 + (true endpoint - corrected endpoint) 2 The corrected start point = provisional start point + correction width, and the corrected end point = provisional end point + correction width.

[0058] For example, if the true starting point of the keyword stuffing is the 10th character and the starting point of the keyword segment predicted by the detector 112(S) is the 8th character, the residual is "+2" in "10-8". The corrector 113(R) is trained to output the residual "+2" using information about the 8th character and its vicinity. If the corrector 113(R) is trained successfully, the "provisionally predicted starting point + corrector output" will match the "true starting point".

[0059] The corrector 113(R) takes the feature value h as an argument, so it may use character information. However, the feature value h may have some information related to position embedded in it by T(v). Therefore, although the corrector 113(R) explicitly uses only the feature value h, which is another expression of character information, the feature value h may contain position information, so the corrector 113(R) ends up using position information as well.

[0060] [B-2] Example of operation The keyword stuffing prediction process in this embodiment will be described with reference to the flowchart (steps S1 to S9) shown in Fig. 9. The process of element #1 in steps S3 to S5 is performed by the detector 112(S), and the process of element #2 in steps S6 to S9 is performed by the corrector 113(R).

[0061] The feature extractor 111(T) extracts a document v=(v1,...,v L), and perform feature extraction on H=(h1,…,h L ) is output (step S1).

[0062] The feature extractor 111(T) divides the image into blocks and outputs H1, . . . , HN (N=(L−K+W) / W) (step S2).

[0063] The detector 112(S) performs the process of the repetition element #1 on Hk (step S3).

[0064] The detector 112(S) detects keyword candidates,

number

[0065] The detector 112(S) determines whether the following (Equation 18) holds (step S5): where t represents a threshold value for determining whether or not a keyword exists.

number

[0066] If (Equation 18) is not true (see the False route in step S5), the process returns to step S3.

[0067] On the other hand, if (Equation 18) is true (see the True route of step S5), the corrector 113(R) calculates the (center, length) notation of the keyword segment.

number

number

[0068] The corrector 113(R) is

number

[0069] The corrector 113(R) calculates the correction width,

number

[0070] The corrector 113(R) performs the correction, calculates the following (Equation 19) (step S9), and outputs (Equation 20), which is a set of the start and end points of the keyword segments. Then, the keyword stuffing prediction process ends.

number

number

[0071] Next, the keyword stuffing prediction process in this embodiment will be described with reference to FIG. 10 (reference numerals F1 to F9).

[0072] The feature extractor 111(T) receives a text input v=(v1,...,v L ) (see symbol F1). The input v is the length L of the string, which may be padded if necessary.

[0073] The feature extractor 111(T) obtains a feature representation h1 (l=1, . . . , L) of each character for the input v (see symbol F2).

[0074] The feature extractor 111(T) aggregates h1 for each block and extracts H k is input to the detector 112(S) (see symbol F3).

[0075] Detector 112(S) is H k Based on this, a three-dimensional vector (p k , c k , l k ) (see symbol F4).k is the probability that keyword stuffing exists in block k, and c k is the coordinate of the center position of keyword stuffing if it exists in block k, and l k is the length of keyword stuffing in block k, if any.

[0076] The detector 112(S) extracts blocks that are determined to contain keyword stuffing, that is, blocks that satisfy the following (Equation 21) (see symbol F5).

number

[0077] The detector 112(S) detects the provisional start position coordinates of the block in which it is determined that keyword stuffing exists.

number

number

[0078] The corrector 113(R)

number

number

[0079] The corrector 113(R)

number

number

[0080] The corrector 113(R) calculates the provisional start and end points.

number

number

number

[0081] [B-3] Hardware configuration example FIG. 11 is a block diagram schematically illustrating an example of the hardware configuration of the information processing device 1. As shown in FIG.

[0082] As shown in FIG. 11, the information processing device 1 includes a CPU 11, a memory 12, a display control device 13, a storage device 14, an input interface (IF) 15, an external recording medium processing device 16, and a communication IF 17.

[0083] The memory 12 is an example of a storage unit, and is illustratively a read-only memory (ROM) and a RAM. A program such as a basic input / output system (BIOS) may be written to the ROM of the memory 12. The software program in the memory 12 may be read and executed by the CPU 11 as appropriate. The RAM of the memory 12 may be used as a temporary storage memory or a working memory.

[0084] The display control device 13 is connected to and controls the display device 131. The display device 131 is a liquid crystal display, an organic light-emitting diode (OLED) display, a cathode ray tube (CRT), an electronic paper display, or the like, and displays various information to the operator of the information processing device 1. The display device 131 may be combined with an input device, and may be, for example, a touch panel.

[0085] The storage device 14 may be, for example, a solid state drive (SSD), a storage class memory (SCM), or a hard disk drive (HDD). The storage device 14 may store a program for executing the keyword stuffing prediction process in the embodiment. The storage device 14 may also store the input v and output y shown in FIG. 4 and the like.

[0086] The input IF 15 may be connected to input devices such as a mouse 151 and a keyboard 152, and may control the input devices such as the mouse 151 and the keyboard 152. The mouse 151 and the keyboard 152 are examples of input devices, and the operator of the information processing device 1 performs various input operations via these input devices.

[0087] The external recording medium processing device 16 is configured so that a recording medium 160 can be attached thereto. The external recording medium processing device 16 is configured so that, when the recording medium 160 is attached, the information recorded on the recording medium 160 can be read. In this example, the recording medium 160 is portable. For example, the recording medium 160 is a non-transitory recording medium such as a flexible disk, an optical disk, a magnetic disk, a magneto-optical disk, or a semiconductor memory.

[0088] The communication IF 17 is an interface that enables communication with an external device.

[0089] The CPU 11 is an example of a processor, and is a processing device that performs various controls and calculations, and realizes various functions by executing an OS and programs loaded into the memory 12. The CPU 11 may be a multiprocessor including multiple CPUs, a multi-core processor having multiple CPU cores, or a configuration having multiple multi-core processors.

[0090] The CPU 11 may function as the feature extractor 111 and the detector 112 shown in FIG. 6 and the like, and may also function as the corrector 113 shown in FIG.

[0091] The device for controlling the overall operation of the information processing device 1 is not limited to the CPU 11, and may be, for example, any one of an MPU, a DSP, an ASIC, a PLD, and an FPGA. The device for controlling the overall operation of the information processing device 1 may also be a combination of two or more of the CPU, MPU, DSP, ASIC, PLD, and FPGA. MPU is an abbreviation for Micro Processing Unit, DSP is an abbreviation for Digital Signal Processor, and ASIC is an abbreviation for Application Specific Integrated Circuit. PLD is an abbreviation for Programmable Logic Device, and FPGA is an abbreviation for Field Programmable Gate Array.

[0092] [C] Effect According to the prediction program, the information processing device 1, and the prediction method in the above-described embodiment, the following effects can be achieved, for example.

[0093] The feature extractor 111 assigns an input first character string to blocks that satisfy a predetermined condition. The detector 112 predicts the probability that keyword stuffing is present in the second character string using the feature values ​​of each character of the second character string included in each block. If the probability is equal to or greater than a predetermined threshold, the detector 112 predicts the center and length of the keyword segment in the second character string.

[0094] This allows keyword stuffing to be detected appropriately. Specifically, the probability of keyword stuffing being present is calculated for each block, and the center and length of the keyword (in other words, keyword segment) to be detected as keyword stuffing are predicted, so that even if the keyword segment has a long character length, the target keyword can be detected without interruption.

[0095] Furthermore, since detection is performed on a character-by-character basis and is language independent, if a multilingual tokenizer is used, the keyword stuffing prediction process in the embodiment can be applied on a token-by-token basis.

[0096] The process of predicting the probability predicts the probability that the true center of the keyword stuffing exists in the second string.

[0097] This allows for more accurate keyword detection by predicting the probability of the center being present.

[0098] a corrector 113 corrects a boundary position indicating at least one of a start position and an end position in the first character string obtained based on the center and length of the predicted keyword segment to a neighboring position indicating a position of a character located near the character at the boundary position; To achieve continuous detection results, global detection, which produces a single prediction result for a certain length unit, is suitable, but because character-by-character accuracy is required for the boundaries of the prediction results, a method that uses only local information near the boundaries is suitable for boundary prediction. Therefore, by using the corrector 113 to correct the prediction results from global detection using local information, continuous detection can be achieved while maintaining prediction accuracy for keyword segment boundaries.

[0099] The process of allocating to blocks is performed such that at least a portion of the second character strings included in adjacent blocks overlap each other.

[0100] This allows for block allocation taking into consideration the correction of boundary positions by the corrector 113.

[0101] The detector 112 is trained to match the centers of the keyword segments to the true centers of the keyword stuffing.

[0102] This allows the accuracy of the detector 112 to be improved.

[0103] [D] Other The disclosed technology is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit of the present embodiment. The configurations and processes of the present embodiment can be selected or combined as needed.

[0104] [E] Supplementary Note The following additional notes are provided regarding the above-described embodiments.

[0105] (Appendix 1) The first character string entered is assigned to a block that satisfies a predetermined condition, predicting the probability that keyword stuffing is present in the second character string using a feature amount of each character of the second character string included in each block and a detector for detecting keyword stuffing; predicting a center and a length of a keyword segment in the second string if the probability is greater than or equal to a predetermined threshold; A prediction program that causes a computer to execute a process.

[0106] (Appendix 2) The process of predicting the probability predicts the probability that a true center of keyword stuffing exists in the second character string. 2. The prediction program according to claim 1,

[0107] (Appendix 3) correcting a boundary position indicating at least one of a start position and an end position in the first character string obtained based on the center and length of the predicted keyword segment to a neighboring position indicating a position of a character located near the character at the boundary position using a corrector that corrects the boundary position; 3. The prediction program according to claim 1 or 2, characterized in that the program causes the computer to execute processing.

[0108] (Appendix 4) the process of allocating to the blocks is performed such that at least a portion of the second character strings included in adjacent blocks overlap with each other. 3. The prediction program according to claim 1 or 2.

[0109] (Appendix 5) training the detector so that the centers of the keyword segments coincide with the true centers of the keyword stuffing; 3. A prediction program according to claim 2, characterized in that the program causes the computer to execute processing.

[0110] (Appendix 6) The first character string entered is assigned to a block that satisfies a predetermined condition, predicting the probability that keyword stuffing is present in the second character string using a feature amount of each character of the second character string included in each block and a detector for detecting keyword stuffing; predicting a center and a length of a keyword segment in the second string if the probability is greater than or equal to a predetermined threshold; An information processing device comprising a processor.

[0111] (Appendix 7) The process of predicting the probability predicts the probability that a true center of keyword stuffing exists in the second character string. 7. The information processing device according to claim 6,

[0112] (Appendix 8) The processor: correcting a boundary position indicating at least one of a start position and an end position in the first character string obtained based on the center and length of the predicted keyword segment to a neighboring position indicating a position of a character located near the character at the boundary position using a corrector that corrects the boundary position; 8. The information processing device according to claim 6 or 7.

[0113] (Appendix 9) the process of allocating to the blocks is performed such that at least a portion of the second character strings included in adjacent blocks overlap with each other. 8. The information processing device according to claim 6 or 7.

[0114] (Appendix 10) The processor: training the detector so that the centers of the keyword segments coincide with the true centers of the keyword stuffing; 8. The information processing device according to claim 7,

[0115] (Appendix 11) The first character string entered is assigned to a block that satisfies a predetermined condition, predicting the probability that keyword stuffing is present in the second character string using a feature amount of each character of the second character string included in each block and a detector for detecting keyword stuffing; predicting a center and a length of a keyword segment in the second string if the probability is greater than or equal to a predetermined threshold; A prediction method characterized in that the processing is performed by a computer.

[0116] (Appendix 12) The process of predicting the probability predicts the probability that a true center of keyword stuffing exists in the second character string. 12. The prediction method of claim 11,

[0117] (Appendix 13) correcting a boundary position indicating at least one of a start position and an end position in the first character string obtained based on the center and length of the predicted keyword segment to a neighboring position indicating a position of a character located near the character at the boundary position using a corrector that corrects the boundary position; 13. The prediction method according to claim 11 or 12, wherein the processing is performed by the computer.

[0118] (Appendix 14) the process of allocating to the blocks is performed such that at least a portion of the second character strings included in adjacent blocks overlap with each other. 13. The prediction method according to claim 11 or 12,

[0119] (Appendix 15) training the detector so that the centers of the keyword segments coincide with the true centers of the keyword stuffing; 13. The prediction method of claim 12, wherein the processing is performed by the computer. [Explanation of symbols]

[0120] 1: Information processing equipment 11: CPU 111: Feature extractor 112: Detector 113:Corrector 12: Memory 13: Display control device 131:Display device 14:Storage device 15: Input IF 151: Mouse 152: Keyboard 16: External recording medium processing device 160: Recording media 17: Communication IF

Claims

1. The first character string input is assigned to a block that satisfies a predetermined condition, predicting the probability that keyword stuffing is present in the second character string using a feature amount of each character of the second character string included in each block and a detector for detecting keyword stuffing; predicting the center and length of a keyword segment in the second string if the probability is greater than or equal to a predetermined threshold; A prediction program that causes a computer to execute a process.

2. The process of predicting the probability predicts the probability that a true center of keyword stuffing exists in the second character string. The prediction program according to claim 1 .

3. correcting a boundary position indicating at least one of a start position and an end position in the first character string obtained based on the center and length of the predicted keyword segment to a neighboring position indicating a position of a character located near the character at the boundary position using a corrector that corrects the boundary position; 3. The prediction program according to claim 1, wherein the program causes a computer to execute the process.

4. the process of allocating to the blocks is performed such that at least a portion of the second character strings included in adjacent blocks overlap with each other.

3. The prediction program according to claim 1 or 2.

5. training the detector so that the centers of the keyword segments coincide with the true centers of the keyword stuffing; The prediction program according to claim 2, wherein the program causes a computer to execute the process.

6. The first character string input is assigned to a block that satisfies a predetermined condition, predicting the probability that keyword stuffing is present in the second character string using a feature amount of each character of the second character string included in each block and a detector for detecting keyword stuffing; predicting the center and length of a keyword segment in the second string if the probability is greater than or equal to a predetermined threshold; An information processing device comprising a processor.

7. The first character string input is assigned to a block that satisfies a predetermined condition, predicting the probability that keyword stuffing is present in the second character string using a feature amount of each character of the second character string included in each block and a detector for detecting keyword stuffing; predicting the center and length of a keyword segment in the second string if the probability is greater than or equal to a predetermined threshold; A prediction method characterized in that the processing is performed by a computer.

Images