DETAILED ACTION
This final office action is responsive to the amendment filed on September 29, 2025. Claims 1-20 are pending. Claims 1, 11, and 20 are independent.
Claim objections are withdrawn in light of applicant’s amendments.
Drawing objections are withdrawn in light of applicant’s replacement drawing.
Specification objections are withdrawn in light of applicant’s amendment to the specification.
Claim rejections under 35 USC §101 are withdrawn in light of applicant’s arguments – see section Response to Arguments below.
Claim rejections under 35 USC §103 are maintained and updated to reflect the amended claims. See sections Claim Rejections – 35 USC §103 and Response to Arguments below.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Drawings
The replacement drawings for Fig. 6 were received on September 29, 2025. These drawings are acceptable.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1 and 4 are rejected under 35 U.S.C. 103 as being unpatentable over Cheng et al. (US20240330716), hereinafter Cheng, in view of Polania Cabrera et al. (US20210110457), hereinafter Cabrera.
Regarding claim 1, Cheng teaches:
A computer implemented method for determining … compatibility using a compatibility system, the method comprising: obtaining a first data set, a second data set, and a first compatibility model; (Cheng, page 0022: “In accordance with another embodiment, the present invention provides a computer-implemented method for efficient ontology matching between a source ontology and a target ontology, the method comprising: filtering out, by means of an adaptive blocking mechanism, non-matching pairs of the source and the target ontology, the method comprising: filtering out, by means of an adaptive blocking mechanism, non-matching pairs of the source and the target ontology, thereby generating an initially unlabeled dataset U of possible matches; selecting, in each iteration of a first learning loop and based on prediction results and uncertainty from a set of initially provided labeling functions of a labeling function, LF, committee” – wherein matching between a source ontology and a target ontology is analogous to measuring compatibility and generating an initially unlabeled data set U encompasses obtaining a first dataset. And paragraph 0030: “LF ensemble component may be implemented within the learning loops that combines voting results
v
i
j
q
1
≤
q
≤
m
from a set of labeling functions
l
f
1
,
l
f
2
,
⋯
,
l
f
m
to predict matching results of all data points of the dataset U” – wherein labeling functions that are used to predict matching results, or the compatibility of two data points, encompasses at least a first compatibility model. And paragraph 0022: “selecting and weighting, in each iteration of the first learning loop, a set of labeling functions out of the LF committee based on their prediction results against the dataset U provided with annotation labels so far and adjusting a weight of each of the selected LFs to produce the prediction results and uncertainty of yet unlabeled data points of dataset U based on the data points of dataset U having already annotated a label.” – wherein the dataset U provided with annotation labels so far is itself a second data set.)
determining, by the compatibility system, a second weight for each instance in the second data set based on a first weight and a weight coefficient of the first compatibility model; (Cheng, paragraph 0060: “selecting one sample based on a number of positive votes and uncertainty given by the LF committee” and paragraph 0063: “adjusting the weights of the selected LFs to produce the prediction results and uncertainty of remaining samples based on the annotated data set” – wherein the remaining samples represents the second data set, and adjusting the weight of the selected LFs to produce the predicted results and uncertainty of the remaining samples is determining a second weight for each instance in the second data set. The LF committee representing the first compatibility model and thus the LFs in the committee would naturally have weight coefficients which are used to determine the second weight for each instance in the second data set and is thus analogous to the weight coefficient of the first compatibility model.)
determining, by the compatibility system, an error instance for each instance in the second data set by applying the first compatibility model to the second data set, the error instance being based on the second weight of each instance in the second data set: (Cheng, paragraph 0022: “selecting and weighting, in each iteration of the first learning loop, a set of labeling functions out of the LF committee based on their prediction results against the dataset U provided with annotation labels so far.” And paragraph 0056: “c) Selecting and weight labeling functions in the fast loop based on both performance metrics and coverage” – Wherein selecting labeling functions based on both performance metrics and coverage encompasses determining error instances in the second data set by applying the first compatibility model to the second data set.)
determining, by the compatibility system, labeling rules based on the error instance for each instance in the second data set; (Cheng, paragraph 0074: “the voting results
v
i
j
q
1
≤
q
≤
m
from the initial set of LFs 224
l
f
1
,
l
f
2
,
⋯
,
l
f
m
, e.g. as provided by the domain expert with low effort, and
v
i
j
q
could be 0 for non-match, 1 for match, or -1 for abstain,” and paragraph 0024: “Further, according to embodiements, the method/system may utilize tunable labeling functions, which can be changed at any time based on performance metrics and coverage” – wherein to “abstain” or classify as “non-match” or “match” are choices encompassing labeling rules. Tunable labeling functions, which can be changed at any time based on performance metrics encompasses determining labeling rules based on the error instances.)
generating, by the compatibility system, a third data set comprising labels for each instance by applying the labeling rules to the first data set… (Cheng, paragraph 0074: “the voting results
v
i
j
q
1
≤
q
≤
m
from the initial set of LFs 224
l
f
1
,
l
f
2
,
⋯
,
l
f
m
, e.g. as provided by the domain expert with low effort, and
v
i
j
q
could be 0 for non-match, 1 for match, or -1 for abstain,” and paragraph 0077: “According to embodiments of the present invention, the LF ensemble 216 is configured to combine the voting results
v
i
j
q
1
≤
q
≤
m
from a set of labeling functions
l
f
1
,
l
f
2
,
⋯
,
l
f
m
to predict the matching results P of all data points in U and also estimate the uncertainty C of the predicted results” – Wherein generating the matching results P of all data points in U using a set of labeling functions encompasses generating a third data set comprising labels for each instance by applying the labeling rules to the first data set.)
generating, by the compatibility system, a second compatibility model based on training the first compatibility model with the third data set; (Cheng, paragraph 0080: “According to embodiments of the present invention, tunable labeling functions may be introduced to create/update new LFs on the fly by automatically tuning some distance-related threshold, based on the latest annotation data A and prediction results P.” – Wherein the labeling functions being updated on the fly…based on the latest annotation data A is analogous to a second compatibility model based on training the first compatibility model with the third data set.)
determining, by the compatibility system, an ensemble compatibility model based on the first compatibility model and the second compatibility model; and (Cheng, paragraph 0030: “According to an embodiment of the invention, an LF ensembler component may be implemented within the learning loops that combines voting results
v
i
j
q
1
≤
q
≤
m
from a set of labeling functions
l
f
1
,
l
f
2
,
⋯
,
l
f
m
to predict matching results of all data points of the dataset U and to estimate an uncertainty of the predicted results.” – Wherein an LF ensemble component is analogous to an ensemble compatibility model based on the first compatibility model and the second compatibility model.)
Cheng does not explicitly teach:
Determining object compatibility
Generating the third data set comprising labels for each instance by applying the labeling rules to the first data set and according to a criterion, the criterion being based on matching product attributes of each instance
determining, by the compatibility system, a product recommendation as output based on applying the ensemble compatibility model to a user selection of a first product as input.
However, Cabrera teaches:
Determining object compatibility (Cabrera, abstract: “Examples disclosed herein are relevant to systems, methods, and other technology for determining furniture compatibility” - Wherein object compatibility encompasses furniture compatibility.)
Generating the third data set comprising labels for each instance by applying the labeling rules to the first data set and according to a criterion, the criterion being based on matching product attributes of each instance (Cabrera, paragraph 0077: “Operation 740 includes returning a subset of the scored items 742 as the selected items 702. In some examples, the scored items 742 will have a range of scores ranging from completely incompatible to completely compatible (e.g., ranging from 0 to 1) with the seed item set 712.” And paragraph 0074: “In an example, the item collection 722 is a subset of furniture items from the item data 130 selected based on the seed item set 712. For example, the item collection 722 can be selected as items in the item data 130 that are of a furniture category other than the one or more furniture categories of the one or more items in the seed item set 712. For instance, where the seed item set 712 includes a coffee table and a couch (e.g., being respectively in the “table” and “couch” categories), the item collection 722 can be furniture items from categories including TV stands, lamps, planters, art, curtains, chairs, or other categories of furniture items that are different from those in the seed item set 712. This can help increase the relevance of the selected items 702 because a user that already has items of one category likely would not want to be recommended items from the same category that are highly compatible the seed item set 712.” – Where the range of scores is analogous to the labeling rules as taught by Cheng, and the subset of furniture items selected based on the seed item set is analogous to the criterion being based on matching product attributes of each instance.)
determining, by the compatibility system, a product recommendation as output based on applying the ensemble compatibility model to a user selection of a first product as input. (Cabrera, Fig. 7, paragraph 0072: “Fig. 7 illustrates a process 700 for selecting one or more items 702 (e.g., to provide as furniture item recommendations) based on a seed item set 712.” paragraph 0019: “The use of one or more GNNs can exploit the underlying relational information between the furniture items to predict compatibility of the given set” and paragraph 0032: “In some examples, the seed item set includes on or more furniture items that are already owned by the user or that are of interest to the user” – The subset of scored items in 740 of fig. 7 represents the product recommendation as the output, one or more GNNs constitutes an ensemble, and the seed item set used as the input in 710 of fig. 7 represents the user selection of a first product as input.)
Cabrera is considered analogous to the claimed invention as it is in the same field of endeavor, machine learning and compatible product pair recommendation systems. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to have modified Cheng, which already teaches a method of assessing compatibility between ontology structures and presenting matches but does not explicitly teach using this method with physical objects, to include the teachings of Cabrera which does teach a method of assessing compatibility between physical objects in order to provide better item compatibility recommendations and is capable of evolving to “provide improved computational accuracy.” (Cabrera, paragraph 0024)
Regarding claim 4, Cheng and Cabrera teach the method of claim 1, as cited above.
Cheng further teaches:
wherein the ensemble compatibility model further comprises a weighted ensemble of preceding compatibility models, wherein each preceding compatibility model includes a weight coefficient. (Cheng, Fig. 2, paragraph 0022: “selecting and weighting, in each iteration of the first learning loop, a set of labeling functions out of the LF committee based on their prediction results against the dataset U provided with annotation labels so far and adjusting a weight of each of the selected LFs to produce the prediction results and uncertainty of yet unlabeled data po9ints of dataset U based on the data points of dataset U having already annotated a label; and executing a second learning loop that automatically creates tuned labeling functions and augments the LF committee with the tuned labeling functions” and paragraph 0078: “it is possible to estimate the precision of the selected LF based on A and then adjust the weight of each selected LF based on its estimated precision for training the LF ensemble model” – In fig. 2 the LF committee 218 consists of initial LFs 224 which encompass preceding compatibility models relative to the new LFs 228 being added and the LF ensemble model encompasses a weighted ensemble of preceding compatibility models.)
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over Cheng in view of Cabrera in view of Zhou et al. (NERO: A Neural Rule Grounding Framework for Label Efficient Relation Extraction), hereinafter Zhou.
Regarding claim 2, Cheng and Cabrera teach the method of claim 1, as cited above.
determining, by the compatibility system, an error rate based on the second data set; (Cheng, paragraph 0022: “selecting and weighting, in each iteration of the first learning loop, a set of labeling functions out of the LF committee based on their predicted results against the dataset U provided with annotation labels so far and adjusting a weight of each of the selected LFs to produce the prediction results and uncertainty of yet unlabeled data points of dataset U based on the data points of dataset U having already annotated a label,” and paragraph 0056: “selecting and weight labeling functions in the fast loop based on performance metrics and coverage,” and paragraph 0078: “First, a heuristic method may be introduced to select a subset of labeling functions out of the LF committee 218 based on A. This way one can exclude some labeling functions that might make negative contributions to the LF ensemble 216.” – wherein the dataset U provided with annotation labels so far is analogous to the second data set, performance metrics encompasses an error rate, and A is equivalent to the dataset U provided with annotation labels so far and determining negative contributions encompasses an error rate.)
determining, by the compatibility system, the weight coefficient for the first compatibility model; and (Cheng, paragraph 0022: “selecting and weighting, in each iteration of the first learning loop, a set of labeling functions out of the LF committee based on their prediction results against the dataset U provided with annotation labels so far and adjusting a weight of each of the selected LFs to produce the prediction results and uncertainty of yet unlabeled data points of dataset U based on the data points of dataset U having already been annotated a label,”- Wherein weighting a set of labeling functions is analogous to determining a weight coefficient for the first compatibility model.)
determining, by the compatibility system, the second weight for each instance of the second data set. (Cheng, paragraph 0060: “selecting one sample based on a number of positive votes and uncertainty given by the LF committee” and paragraph 0063: “adjusting the weights of the selected LFs to produce the prediction results and uncertainty of remaining samples based on the annotated data set” – wherein the remaining samples represents the second data set, and adjusting the weight of the selected LFs to produce the predicted results and uncertainty of the remaining samples is determining a second weight for each instance in the second data set.)
Cheng and Cabrera do not explicitly teach:
determining, by the compatibility system, the first weight for each instance of the second data set;
However, Zhou teaches:
determining, by the compatibility system, the first weight for each instance of the second data set; (Zhou, section 4.1 Parameter Learning of NERO, paragraph 4: “Similarly, we sample
B
u
from
S
unmatched
, and generate the pseudo-labels with the entire rule set
P
, and calculating normalized weights for
L
unmatched
by:
PNG
media_image1.png
130
562
media_image1.png
Greyscale
The normalized instance weights ensure the scale of
L
unmatched
to be stable across different steps” – Wherein to calculate weights for [each sentence in]
L
unmatched
is analogous to determining a first weight for each instance in the second data set.)
Zhou is considered analogous to the claimed invention as it is in the same field of endeavor, machine learning and methods of matching data points using labeling rules. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to have modified Cheng and Cabrera, which already teaches a self-labeling data programming framework but does not explicitly teach generating the first weights based off each instance of the second dataset, to include the teachings of Zhou which does teach generating the first weights based off each instance of the second dataset in order to improve accuracy by reformatting the labeling rules to better fit the training data. (Zhou, section 3.5, page 5, col. 1, paragraph 3)
Claim rejected under 35 U.S.C. 103 as being unpatentable over Cheng in view of Cabrera in view of Zhou in view of Varma et al. (Snuba: Automating Week Supervision to Label Training Data), hereinafter Varma.
Regarding claim 3, Cheng, Cabrera, and Zhou teach the method of claim 2, as cited above.
Cheng, Cabrera, and Zhou do not explicitly teach:
wherein the instances in the second data set with higher second weights are treated as large error instances, and wherein the labeling rules are based on the large error instances in the second data set.
However, Varma teaches:
wherein the instances in the second data set with higher second weights are treated as large error instances, and wherein the labeling rules are based on the large error instances in the second data set. (Varma, section 3.3, page 228, col. 1, paragraphs 3-4: “These probabilistic labels also represent how confident the label aggregator is about the assigned label… We also compare to a weighted feedback approach in which the weights are the inverse of the label confidence
w
v
=
1
2
-
y
~
-
1
2
normalized across all datapoints,” – Wherein label confidence encompasses error. Therefore, instances with higher second weights due to the inverse relationship to the confidence, necessarily have a large error, these instances being assigned more importance due to the weighted feedback approach informing the label heuristics, or labeling rules, of Varma. As in Cheng and Zhou, Varma performs an iterative labeling process indicating that these label confidence scores encompass second weights.)
Varma is analogous to the claimed invention as both are from the same field of endeavor, that is, automatic data labeling through weak supervision. Cheng, Cabrera, and Zhou teaches a method of assigning labels uncertainty values but does not explicitly reference these labels as directly informing the labeling functions. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, the self-labeling data programming framework of Cheng to incorporate the confidence weighting of Varma. The motivation to do so is to iteratively improve the predictive label accuracy by applying the updated labeling functions to lower confidence data points (Varma, 3.3 Verifier, pp. 229, col. 1, paragraph 3; "P[y* = 1] close to 0.5 represent datapoints with low confidence, which can result from scenarios with low accuracy heuristics labeling that datapoint, or multiple heuristics with similar accuracies disagreeing on the label for that datapoint. Since Snuba generates a new heuristic at each iteration, we want the new heuristic to assign labels to the subset that currently has low confidence labels.").
Claims 5-11 and 14-20 are rejected under 35 U.S.C. 103 as being unpatentable over Cheng in view of Cabrera in view of Forsyth et al. (US20200311798), hereinafter Forsyth.
Regarding claim 5, Cheng and Cabrera teach the method of claim 1, as cited above.
Cheng and Cabrera do not explicitly teach:
wherein the first data set includes unlabeled product pair data, the unlabeled product pair data includes an anchor product from a product category and randomly sampled second products from the product category, wherein the anchor product comprises the first product selectable by the user.
However, Forsyth teaches:
wherein the first data set includes unlabeled product pair data, the unlabeled product pair data includes an anchor product from a product category and randomly sampled second products from the product category, (Forsyth, paragraph 0069: “A triplet is a set of images
x
i
(
u
)
,
x
j
(
v
)
,
x
k
(
v
)
with the following relationship: an anchor image
x
i
,
is of some type u, and both
x
j
and
x
k
are of a different type v. The pair
x
i
,
x
i
is compatible, meaning that the two items appear together in an outfit, while
x
k
is a randomly sampled item of the same type as
x
j
that has not been seen in an outfit with
x
i
. In the present example, the top (associated with the first image 410A) can be viewed as the anchor image, the pants (associated with the second image 410B) as the compatible item, and a random new pants can be viewed as the randomly sampled item,
x
k
,” and Fig. 4B – Wherein an anchor image
x
i
and
x
j
correspond to anchor product[s]. In this case,
x
j
is considered an anchor product as its type determines the sampling of
x
k
. Additionally,
x
k
encompasses randomly sampled second products from the same product category, or type, as the anchor product
x
j
. Fig. 4B indicates that the triplet of images, 410A-C, is selected before being passed to the image embedder 460 which indicates that it is unlabeled.)
wherein the anchor product comprises the first product selectable by the user. (Forsyth, paragraph 0059: “In associated embodiments, the visual semantic embedder 124 scrapes metadata from or associated with each of the images 410A, 410B, and 410C selected from the online site(s) to retrieve text that is descriptive of each of the images,” – Thereby indicating that the “pants” corresponding to
x
j
are selectable by the user.)
Forsyth is analogous to the claimed invention as both are from the same field of endeavor, that is, compatible product pair recommendation systems. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to employ the self-labeling data programming framework of Chen and Cabrera in the product pair recommendation system of Forsyth. The motivation to do so is to reduce the time and effort required in labeling vast amounts of datasets required for such systems (Forsyth, [0040]; "For training fashion characteristics extraction model, acquiring accurate and complete labeling is difficult. Some works labeled the dataset through human annotation with a limited number of learning targets, while others learned styles through unsupervised approaches. Many works identified the challenge of incomplete and noisy labeling introduced by inferring training labels from metadata related to fashion products.").
Regarding claim 6, Cheng and Cabrera teach the method of claim 1, as cited above.
Cheng further teaches:
wherein the second data set includes labeled … pair data from a preceding iteration. (Cheng, paragraph 0022: “selecting and weighting in each iteration of the first learning loop, a set of labeling functions out of the LF committee based on their prediction results against the dataset U provided with annotation labels so far,” – Wherein the dataset U provided with annotation labels so far” encompasses the second data set including ontology pair data from the preceding iteration.)
Cheng and Cabrera do not explicitly teach product pair data
However, Forsyth teaches this limitation (Forsyth, paragraph 0069: “A triplet is a set of images
x
i
(
u
)
,
x
j
(
v
)
,
x
k
(
v
)
with the following relationship: an anchor image
x
i
,
is of some type u, and both
x
j
and
x
k
are of a different type v. The pair
x
i
,
x
i
is compatible, meaning that the two items appear together in an outfit, while
x
k
is a randomly sampled item of the same type as
x
j
that has not been seen in an outfit with
x
i
.” – Wherein a triplet comprising a pair
x
i
,
x
i
of clothing items encompasses product pair data.)
Forsyth is analogous to the claimed invention as both are from the same field of endeavor, that is, compatible product pair recommendation systems. Cheng and Cabrera teaches a method of assessing compatibility between ontology structures and presenting matches, but does not teach such a method with physical objects. Forsyth teaches this limitation. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to employ the self-labeling data programming framework of Cheng and Cabrera in the product pair recommendation system of Forsyth. The motivation to do so is to reduce the time and effort required in labeling vast amounts of datasets required for such systems (Forsyth, [0040]; "For training fashion characteristics extraction model, acquiring accurate and complete labeling is difficult. Some works labeled the dataset through human annotation with a limited number of learning targets, while others learned styles through unsupervised approaches. Many works identified the challenge of incomplete and noisy labeling introduced by inferring training labels from metadata related to fashion products.").
Regarding claim 7, Cheng and Cabrera teach the method of claim 1, as cited above.
Cheng further teaches:
wherein the third data set comprises labeled …pair data based on the labeling rules being applied to the first data set. (Cheng, paragraph 0077: “According to embodiments of the present invention, the LF ensembler 216 is configured to combine the voting results
v
i
j
q
1
≤
q
≤
m
from a set of labeling functions
l
f
1
,
l
f
2
,
⋯
,
l
f
m
to predict the matching results P of all data points in U and also estimate the uncertainty C of the predicted results,” – Wherein the matching results P comprises the third data set based on the labeling rules, from the set of labeling functions being applied to the first data set, U.)
Cheng and Cabrera do not explicitly teach product pair data
However, Forsyth teaches this limitation (Forsyth, paragraph 0069: “A triplet is a set of images
x
i
(
u
)
,
x
j
(
v
)
,
x
k
(
v
)
with the following relationship: an anchor image
x
i
,
is of some type u, and both
x
j
and
x
k
are of a different type v. The pair
x
i
,
x
i
is compatible, meaning that the two items appear together in an outfit, while
x
k
is a randomly sampled item of the same type as
x
j
that has not been seen in an outfit with
x
i
.” – Wherein a triplet comprising a pair
x
i
,
x
i
of clothing items encompasses product pair data.)
Forsyth is analogous to the claimed invention as both are from the same field of endeavor, that is, compatible product pair recommendation systems. Cheng and Cabrera teaches a method of assessing compatibility between ontology structures and presenting matches, but does not teach such a method with physical objects. Forsyth teaches this limitation. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to employ the self-labeling data programming framework of Cheng and Cabrera in the product pair recommendation system of Forsyth. The motivation to do so is to reduce the time and effort required in labeling vast amounts of datasets required for such systems (Forsyth, [0040]; "For training fashion characteristics extraction model, acquiring accurate and complete labeling is difficult. Some works labeled the dataset through human annotation with a limited number of learning targets, while others learned styles through unsupervised approaches. Many works identified the challenge of incomplete and noisy labeling introduced by inferring training labels from metadata related to fashion products.").
Regarding claim 8, Cheng, Cabrera, and Forsyth teach the method of claim 7, as cited above.
Cheng further teaches:
wherein applying the labeling rules to the first data set comprises providing an … instance for each … pair indicating a compatibility of the … pair in the first data set. (Cheng, abstract: “A method for ontology matching between a source and a target filters out non-matching pairs of the source and the target, to generate a dataset of possible matches” and paragraph 0074: “the voting results
v
i
j
q
1
≤
q
≤
m
from the initial set of LFs 224
l
f
1
,
l
f
2
,
⋯
,
l
f
m
, e.g. as provided by the domain expert with low effort, and
v
i
j
q
could be 0 for non-match, 1 for match, or -1 for abstain,” – Wherein to assign a label indicating a non-match or match is equivalent to applying labeling rules providing an instance for each pair of ontologies indicating a compatibility of the pair.)
Cheng and Cabrera do not explicitly teach an object instance, products, or product pairs.
However, Forsyth further teaches these limitations (Forsyth, paragraph 0069: “A triplet is a set of images
x
i
(
u
)
,
x
j
(
v
)
,
x
k
(
v
)
with the following relationship: an anchor image
x
i
,
is of some type u, and both
x
j
and
x
k
are of a different type v. The pair
x
i
,
x
i
is compatible, meaning that the two items appear together in an outfit, while
x
k
is a randomly sampled item of the same type as
x
j
that has not been seen in an outfit with
x
i
.” – Wherein a triplet comprising a pair
x
i
,
x
i
of clothing items encompasses an object instance of product pair data.)
Regarding claim 9, Cheng and Cabrera teach the method of claim 1, as cited above.
Cheng further teaches:
wherein the labeling rules comprise being based on shared first attributes and second attributes between candidates in each … pair. (Cheng, paragraphs 0033-0035: “In this context, it may be provided that each tunable labeling function relies on a tunable threshold parameter and a predefined similarity feature to decide whether a candidate
d
i
j
in U is a match or not based on a predefined logic. According to embodiments of the invention, the predefined similarity feature may be configured to measure similarity based on any of the following methods: (1) using a pre-trained sentence transformer model measuring a semantic similarity of string-based attributes of the classes
e
i
,
e
j
,” – Wherein a semantic similarity of string based attributes encompasses shared first attributes and second attributes. That the labeling function relies on [this] … similarity feature indicates that the labeling rules are based on these attributes while the candidate
d
i
j
in U matching or not is checking the candidates between each pair in the data.)
Cheng and Cabrera do not explicitly teach that the labeling rules are between products in each product pair.
However, Forsyth teaches this limitation (Forsyth, paragraph 0069: “A triplet is a set of images
x
i
(
u
)
,
x
j
(
v
)
,
x
k
(
v
)
with the following relationship: an anchor image
x
i
,
is of some type u, and both
x
j
and
x
k
are of a different type v. The pair
x
i
,
x
i
is compatible, meaning that the two items appear together in an outfit, while
x
k
is a randomly sampled item of the same type as
x
j
that has not been seen in an outfit with
x
i
.” – Wherein a triplet comprising a pair
x
i
,
x
i
of clothing items encompasses product pair data.)
Forsyth is analogous to the claimed invention as both are from the same field of endeavor, that is, compatible product pair recommendation systems. Cheng and Cabrera teaches a method of assessing compatibility between ontology structures and presenting matches, but does not teach such a method with physical objects. Forsyth teaches this limitation. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to employ the self-labeling data programming framework of Cheng and Cabrera in the product pair recommendation system of Forsyth. The motivation to do so is to reduce the time and effort required in labeling vast amounts of datasets required for such systems (Forsyth, [0040]; "For training fashion characteristics extraction model, acquiring accurate and complete labeling is difficult. Some works labeled the dataset through human annotation with a limited number of learning targets, while others learned styles through unsupervised approaches. Many works identified the challenge of incomplete and noisy labeling introduced by inferring training labels from metadata related to fashion products.").
Regarding claim 10, Cheng, Cabrera, and Forsyth teach the method of claim 9, as cited above.
Cheng further teaches:
wherein the first attributes comprise structured attributes; and (Cheng, paragraph 0041: “Each ontology defines a conceptualization of a domain with classes (representing domain concepts), literal values (such as comments and labels), individuals (instances of classes) and properties. Properties can define relations between classes/instances (object properties) and relations between a class/instance and a literal (data property), thus defining the attributes. Classes can have super and sub-class relations, defining a hierarchical structure,” – Wherein the representation of a class, or an ontology, encompasses defining a hierarchical structure. Therefore, these attributes are by nature structured.)
Cheng and Cabrera do not explicitly teach:
wherein the second attributes comprise unstructured attributes.
However, Forsyth further teaches:
wherein the second attributes comprise unstructured attributes. (Forsyth, paragraph 0071: “In addition to scoring type-dependent compatibility, the search engine server 120 also trains the visual semantic embedder 124 using features of text descriptions accompanying each item, which regularizes the general embedding and the learned compatibility and similarity relationships,” – Wherein text descriptions encompasses unstructured attributes according with the explanation provided at paragraph 0041 of the instant specification, “As described herein, the term “unstructured product attribute,” and “unstructured attribute” may refer to unlabeled data (e.g., product overviews, product descriptions, etc.).”)
Regarding claim 11, Claim 11 has all the same limitations of claims 1 and 9, which are taught by Cheng, Cabrera, and Forsyth – see claims 1 and 9 above.
Cheng and Cabrera do not explicitly teach:
A computer-implemented method for determining object compatibility using a neural network or by a neural network.
However, Forsyth further teaches this limitation. (Forsyth, abstract: “A processing device, coupled to the communication interface, is to: execute a neural network (NN) regressor model on the first image to identify a plurality of second items that are similar to and compatible with the item depicted in the first image, wherein a set of images correspond to the plurality of second items.”)
Regarding claim 14, Cheng, Cabrera, and Forsyth teach the method of claim 11, as cited above.
Claim 14 additionally has the same limitations of claim 4 which are taught by Cheng and Cabrera – see claim 4 above.
Regarding claim 15, Cheng, Cabrera, and Forsyth teach the method of claim 11, as cited above.
Claim 15 additionally has the same limitations of claim 5 which are taught by Cheng, Cabrera, and Forsyth – see claim 5 above.
Regarding claim 16, Cheng, Cabrera, and Forsyth teach the method of claim 11, as cited above.
Claim 16 additionally has the same limitations of claim 6 which are taught by Cheng, Cabrera, and Forsyth – see claim 6 above.
Regarding claim 17, Cheng, Cabrera, and Forsyth teach the method of claim 11, as cited above.
Claim 17 additionally has the same limitations of claim 7 which are taught by Cheng, Cabrera, and Forsyth – see claim 7 above.
Regarding claim 18, Cheng, Cabrera, and Forsyth teach the method of claim 17, as cited above.
Claim 18 additionally has the same limitations of claim 8 which are taught by Cheng, Cabrera, and Forsyth – see claim 8 above.
Regarding claim 19, Cheng, Cabrera, and Forsyth teach the method of claim 11, as cited above.
Claim 19 additionally ahs the same limitations of claim 10 which are taught by Cheng, Cabrera, and Forsyth – see claim 10 above.
Regarding claim 20, Claim 20 is a system having the same limitations of claim 11 which are taught by Cheng, Cabrera, and Forsyth – see claim 11 above.
While Cheng teaches that the method is computer-implemented, it does not explicitly teach:
the system comprising: a processor; and a non-transitory, computer-readable memory storing instructions that, when executed by the processor, cause the system to perform a method comprising:
However, Cabrera further teaches:
the system comprising: a processor; and a non-transitory, computer-readable memory storing instructions that, when executed by the processor, cause the system to perform a method comprising: (Cabrera, paragraph 0007: “An example system comprises a processor and a memory communicatively coupled to the processor. The memory stores instructions that, when executed by the processor, cause the system to:”)
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Cheng in view of Cabrera in view of Forsyth in view of Zhou.
Regarding claim 12, Cheng, Cabrera, and Forsyth teach the method of claim 11, as cited above.
Claim 12 additionally has the same limitations of claim 2 which are taught by Cheng, Cabrera, and Zhou – see claim 2 above.
Claim 1 rejected under 35 U.S.C. 103 as being unpatentable over Cheng in view of Cabrera in view of Forsyth in view of Zhou in view of Varma.
Regarding claim 13, Cheng, Cabrera, Forsyth, and Zhou teach the method of claim 12, as cited above.
Claim 13 additionally has the same limitations of claim 3 which are taught by Cheng, Cabrera, Zhou, and Varma – see claim 3 above.
Response to Arguments
Applicant’s arguments, see 13 of Applicant's Remarks, with respect to the rejection of claims 1-20 under 35 USC 101 have been fully considered and are persuasive. In particular, applicant argues that the claimed invention represents an improvement to the functionality of compatibility systems, in particular “the ‘ensemble compatibility model’ is utilized to provide product recommendations that are based on the ‘second compatibility model,’ which ‘target[s large] error instances to provide candidate labeling rules to complement the current rule set R, and to suppress the noise in the initial weak data set and adaptively improve the ECM.” The rejection of claims 1-20 under 35 USC 101 has been withdrawn.
Applicant's arguments regarding claim rejections under 35 USC §103 have been fully considered but they are not persuasive. First, examiner notes that while the previous examiner noted that Cheng did not teach the second weight for each instance of the second data set, upon further consideration, in light of applicant’s amendments, Cheng does teach this feature. At least in paragraphs 0031, 0063, and 0078 Cheng discusses adjusting weights of the LFs based on the sample set pulled from the data set U which would be analogous to the second data set and the adjusted weights are analogous to the second weight for each instance in the second data set based on the first weight and a weight coefficient of the first compatibility model as the LFs are weighted. In regards to applicant’s arguments on page 17 that Cheng does not teach the error instance of each instance in the second data set, the examiner disagrees. The selecting and weighting based on the performance metrics and coverage is the error instance and this loop being run on the labeling functions “against the dataset U provided with the annotation labels so far” indicates that “the dataset U provided with the annotation labels so far” is the second dataset while it being done “against” that would be each instance in the second dataset.
Applicant’s arguments on page 17 of Applicant’s Remarks regarding the combination of Cheng and Zhou is moot since, as noted above, Cheng is relied upon to teach the amended limitation of the second weights which is then being used for the modifications in claim 1 and Zhou is not relied upon for the rejection of claim 1.
Claim rejections of claims 1-20 under 35 USC §103 are maintained and updated to reflect the currently amended claims. See section Claim Rejections – 35 USC §103 above.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JACQUELINE MEYER whose telephone number is (703)756-5676. The examiner can normally be reached M-F 8:00 am - 4:30 pm EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Tamara Kyle can be reached at 571-272-4241. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/J.C.M./Examiner, Art Unit 2144
/TAMARA T KYLE/Supervisory Patent Examiner, Art Unit 2144
Prosecution Insights